What You'll Do
Build machine learning models for robotics control end-to-end: data curation, careful evaluation, model architecture, training/inference stacks, rigorous experiments
Collaborate with simulation and real-world robotics teams to curate high-quality, diverse, and large-scale datasets
Curate the world’s best Internet-scale datasets for embodied perception and first-person robot video generation
Design new generative simulation techniques to expand simulation data scale and diversity, training and evaluating generative models of 3D objects and environments, and language/code models to generate tasks and reward functions
Collaborate with a team of driven individuals committed to building general-purpose Physical AI
What You'll Bring
Passion for your craft and demonstrated excellence in full-stack foundation model research and engineering
Exceptional ownership and initiative—finding and solving problems independently
Extensive experience pioneering new machine learning ideas or refining existing methods, supported by first-author publications or impactful projects (5+ years)
A relentless commitment to data and code quality, rigorous evaluation, and meticulous attention to detail
Production-level expertise in modern Python
Bonus: Experience with Vision-Language-Action models for robotics or web agents, embodied perception, or video generation
What You’ll Do
Develop LM/VLM-powered agents that autonomously interface with code, GUI, web tools, and design software to drive generative simulation
Build agentic systems that synthesize high-quality, diverse physical data by orchestrating simulation, generative models, and interactive tools
Build data pipelines to curate, generate, and preprocess datasets used for training and evaluating language- and vision-language-based agents
Collaborate with simulation and infrastructure teams to co-develop APIs, GUIs, and services that enable seamless interaction with physics engines and simulation environments
Drive innovation by staying at the forefront of LLM agents, autonomous tool use, and generative simulation — and translating cutting-edge research into robust, production-ready systems
Contribute to the creation of a generative simulation stack that empowers downstream applications in robotics, embodied AI, and physical reasoning
What You’ll Bring
A bold and imaginative vision for a next-generation paradigm of physical data synthesis— combining simulation, generative models, and autonomous agents
Deep curiosity and strong technical ownership, with a track record of driving complex, open-ended projects from concept to implementation
Experience with (multimodal) large language models, generative AI tools, agentic software design, GUI automation, program synthesis
Passion for inventing creative solutions at the edge of AI, simulation, and physical reasoning, with an eagerness to build systems that generate high-fidelity, useful data for embodied intelligence
Bonus: Familiarity with interactive design tools (e.g., CAD software, game engines) or simulation environments (e.g., physics engines, robotics simulators)
What You’ll Do
Develop and optimize a learning-based robotic manipulation control stack
Design and maintain a teleoperation system with smooth, precise motion and low latency
Train robotic policies for manipulation and locomotion with reinforcement learning and imitation learning
Deploy robotic policies and diagnose latency or bottlenecks in the control pipeline
Analyze and minimize the sim-to-real gap by co-optimizing simulation and real-world robot behavior
Collaborate with a team of driven individuals committed to building general-purpose Physical AI
What You’ll Bring
Passion for your craft and demonstrated excellence in robotics engineering
Exceptional ownership and initiative—finding and solving problems independently
Focus, attention to detail, patience, and a methodical approach to complex tasks
Production-level expertise in modern Python or C++
Extensive experience building and deploying real-world learning-based robotic systems (5+ years)
Bonus: Hands-on experience troubleshooting and maintaining robotic systems across mechanical, electrical, and software components
What You’ll Do
Design, build, and maintain large-scale data pipelines (batch and streaming) for robotics foundation model training and evaluation at petabyte scale
Own core data infrastructure: data model, storage systems, ingestion pipelines, transformation frameworks, and orchestration layers
Standardize data models and unify processing pipelines across real-world teleoperation and synthetic simulation datasets
Collaborate with a team of driven individuals committed to building general-purpose Physical AI
What You’ll Bring
Excellent software engineering skills (Python, Go, or similar)
Extensive experience designing, building, and maintaining large-scale data pipelines (8+ years)
Deep understanding of distributed systems (Spark, Kafka, or similar)
Extensive experience with data storage technologies (data lakes, warehouses, object stores like S3)
Experience running and maintaining production-grade infrastructure (Kubernetes, Terraform)
Bonus: Experience supporting AI systems, in particular embodied AI like self-driving
Build core simulation-centered infrastructure for robotics model development, evaluation, and iteration.
Design scalable systems for creating, running, and managing diverse simulation workloads across tasks, environments, and robots.
Develop reusable environment frameworks, task interfaces, and tooling that make simulation workflows more reliable and observable.
Work with the policy training team to bridge the sim-to-real gap
Develop a real-to-sim pipeline in collaboration with the asset generation team
Collaborate with a team of driven individuals committed to building general-purpose Physical AI
Passion for your craft and demonstrated excellence in physics simulation engineering
Exceptional ownership and initiative: finding and solving problems independently
Extensive experience in physics simulation, with a preference for expertise in rigid body simulation and FEM
Insights and requests from a robotic researcher/engineer’s perspective for improving user experience
Proficiency in CUDA and GPU programming
Built closed-loop or log-replay evaluation at scale.
Experience with robotics simulation tools such as genesis-world, Isaac Sim, MuJoCo, Newton, Drake, Gazebo, or similar platforms.
Experience with synthetic data, scene generation, asset pipelines, sim-to-real, or robotics benchmarks.
Experience with robot manipulation, mobile manipulation, humanoids, or embodied AI systems.
Quadrants is an open-source translation layer that turns ordinary Python into optimized machine code at runtime, across CPU (Arm64, x86) and GPU (AMDGPU, CUDA, Metal, Vulkan) backends. CUDA is the primary target, with x86 the strong runner-up, especially for real-time simulation. The others drive adoption and support fast prototyping for research.
The ambition is explicit: build the best “Python compiler” in the world for physics and performance-critical numerical computing. We are fully committed to it, and we intend to beat leading alternatives like Nvidia Warp, Numba, or Triton.
It’s a standalone project with its own roadmap, built in the open. Genesis, our physics simulator, is its first and top-priority user, the source of its hardest workloads and its benchmarks. But Quadrants is general-purpose infrastructure for anyone writing performance-critical numerical Python, not an internal tool that happens to be public.
The bet: high-performance code should feel natural to write. Engineers write plain Python in a NumPy or PyTorch style, and Quadrants makes it fast.
You own Quadrants end to end: the API, the IR and lowering, the optimization passes, the numerical primitives, and the roadmap.
Our ultimate objective: engineers write for readability and maintainability, never for performance. They write the version of an algorithm they are most comfortable with, and Quadrants rebuilds it into kernels that run as if an expert had hand-tuned them. Performance is the compiler’s job, not theirs. Getting there is a frontier problem, part research and part engineering, and we intend to define the state of the art on it, relentlessly pushing the limit further.
It calls for two things at once: the low-level GPU and compiler skill to make a single kernel optimal, and the ability to keep the whole system coherent as it grows.
The core tradeoff is settled: usability comes first. When usability and performance collide, usability wins, unless the cost is severe. We will limit what the system supports to go fast (static shapes and fixed-size allocation baked into kernels, say), but never the convenience of the API itself. Experts who want maximum performance and full control over the generated code get lower-level escape hatches, down to inline assembly, C, or PTX.
API & authoring experience. The surface that physics and numerical-computing engineers write against, and, as an open-source library, the product’s face to the world. Idiomatic, natural to Python users, never a separate language bolted onto Python. Clean decorators, legible constraints, sharp errors, real escape hatches.
IR, lowering & backends. A stable IR that lowers to multiple backends, with room to grow and the door kept open for differentiable simulation. Today the computation graph is expressed through Quadrants’ own constructs. Recovering it automatically from ordinary Python is hard, and a long-term goal rather than today’s reality.
Optimization. Turn naive code into strong execution plans: vectorization, register and shared-memory usage, barriers and atomics, cooperative threading, dead-buffer and redundant-traffic elimination, dropping needless synchronization. Move low-level optimization off users and into the platform.
Primitives. Cholesky, eigendecomposition, matmul, reductions, solver kernels. Each owned with its correctness, benchmarks, and regression tracking.
Watching Genesis and the compiler together. Hunt for performance wins on both sides, in Genesis and in Quadrants itself. Propose API changes that cut duplication, improve readability, and reduce platform-specific special-casing. Less to maintain, more that simply works.
The project as a product. Benchmark real workloads, track performance over time, catch regressions early. Stay close to users (Genesis first), find friction and missing abstractions, drive adoption, set direction.
You reason about whole systems and have taste for where complexity should live. You measure before you optimize, and you reach for memory traffic and launch overhead before FLOPs. You have opinions about API design. You write tests and benchmarks as you go, not afterward.
The ground that judgment stands on:
GPU. What optimal looks like and why: access patterns, registers, shared memory, barriers, atomics, threading. CPU intuition a plus.
Python & API taste. Idiomatic Python. A feel for what’s natural to Python users. APIs that free users rather than constrain them.
Compilers. The path from authoring to kernels (IR, lowering, optimization), weighed against maintainability.
And how you work: proactive and largely self-directed. You fetch context across the company instead of waiting for it. You’re comfortable talking to the engineers who use what you build, and at ease designing in the open for users you’ll never meet. Ready to grow into leading a small team (a couple of people, near term).
People adopt Quadrants from outside Genesis. Engineers write clearer code with less boilerplate. Kernels keep getting faster. Performance is tracked, not guessed. The architecture gets easier to extend, not harder.
How you work is up to you. What matters is that your ownership grows over time: blind spots shrink, the parts you can vouch for expand, technical debt goes down, the core stays sound.
That takes treating understanding as real work, not overhead. You read and challenge the codebase instead of just stacking features on it. You take nothing you did not write for granted. You watch the competition and what’s coming next, and you bring it in and ship it.
Use AI agents if they help. They are productivity tools, not code owners. You are.
","descriptionPlain":"WHAT QUADRANTS IS\n\nQuadrants is an open-source translation layer that turns ordinary Python into optimized machine code at runtime, across CPU (Arm64, x86) and GPU (AMDGPU, CUDA, Metal, Vulkan) backends. CUDA is the primary target, with x86 the strong runner-up, especially for real-time simulation. The others drive adoption and support fast prototyping for research.\n\nThe ambition is explicit: build the best “Python compiler” in the world for physics and performance-critical numerical computing. We are fully committed to it, and we intend to beat leading alternatives like Nvidia Warp, Numba, or Triton.\n\nIt’s a standalone project with its own roadmap, built in the open. Genesis, our physics simulator, is its first and top-priority user, the source of its hardest workloads and its benchmarks. But Quadrants is general-purpose infrastructure for anyone writing performance-critical numerical Python, not an internal tool that happens to be public.\n\nThe bet: high-performance code should feel natural to write. Engineers write plain Python in a NumPy or PyTorch style, and Quadrants makes it fast.\n\n\n\n\nTHE ROLE\n\nYou own Quadrants end to end: the API, the IR and lowering, the optimization passes, the numerical primitives, and the roadmap.\n\nOur ultimate objective: engineers write for readability and maintainability, never for performance. They write the version of an algorithm they are most comfortable with, and Quadrants rebuilds it into kernels that run as if an expert had hand-tuned them. Performance is the compiler’s job, not theirs. Getting there is a frontier problem, part research and part engineering, and we intend to define the state of the art on it, relentlessly pushing the limit further.\n\nIt calls for two things at once: the low-level GPU and compiler skill to make a single kernel optimal, and the ability to keep the whole system coherent as it grows.\n\nThe core tradeoff is settled: usability comes first. When usability and performance collide, usability wins, unless the cost is severe. We will limit what the system supports to go fast (static shapes and fixed-size allocation baked into kernels, say), but never the convenience of the API itself. Experts who want maximum performance and full control over the generated code get lower-level escape hatches, down to inline assembly, C, or PTX.\n\n\n\n\nWHAT YOU’LL OWN\n\n - API & authoring experience. The surface that physics and numerical-computing engineers write against, and, as an open-source library, the product’s face to the world. Idiomatic, natural to Python users, never a separate language bolted onto Python. Clean decorators, legible constraints, sharp errors, real escape hatches.\n\n - IR, lowering & backends. A stable IR that lowers to multiple backends, with room to grow and the door kept open for differentiable simulation. Today the computation graph is expressed through Quadrants’ own constructs. Recovering it automatically from ordinary Python is hard, and a long-term goal rather than today’s reality.\n\n - Optimization. Turn naive code into strong execution plans: vectorization, register and shared-memory usage, barriers and atomics, cooperative threading, dead-buffer and redundant-traffic elimination, dropping needless synchronization. Move low-level optimization off users and into the platform.\n\n - Primitives. Cholesky, eigendecomposition, matmul, reductions, solver kernels. Each owned with its correctness, benchmarks, and regression tracking.\n\n - Watching Genesis and the compiler together. Hunt for performance wins on both sides, in Genesis and in Quadrants itself. Propose API changes that cut duplication, improve readability, and reduce platform-specific special-casing. Less to maintain, more that simply works.\n\n - The project as a product. Benchmark real workloads, track performance over time, catch regressions early. Stay close to users (Genesis first), find friction and missing abstractions, drive adoption, set direction.\n\n\nWHO YOU ARE\n\nYou reason about whole systems and have taste for where complexity should live. You measure before you optimize, and you reach for memory traffic and launch overhead before FLOPs. You have opinions about API design. You write tests and benchmarks as you go, not afterward.\n\nThe ground that judgment stands on:\n\n - GPU. What optimal looks like and why: access patterns, registers, shared memory, barriers, atomics, threading. CPU intuition a plus.\n\n - Python & API taste. Idiomatic Python. A feel for what’s natural to Python users. APIs that free users rather than constrain them.\n\n - Compilers. The path from authoring to kernels (IR, lowering, optimization), weighed against maintainability.\n\nAnd how you work: proactive and largely self-directed. You fetch context across the company instead of waiting for it. You’re comfortable talking to the engineers who use what you build, and at ease designing in the open for users you’ll never meet. Ready to grow into leading a small team (a couple of people, near term).\n\n\n\n\nWHAT GREAT LOOKS LIKE\n\nPeople adopt Quadrants from outside Genesis. Engineers write clearer code with less boilerplate. Kernels keep getting faster. Performance is tracked, not guessed. The architecture gets easier to extend, not harder.\n\n\n\n\nWHAT WE’RE UNCOMPROMISING ABOUT\n\nHow you work is up to you. What matters is that your ownership grows over time: blind spots shrink, the parts you can vouch for expand, technical debt goes down, the core stays sound.\n\nThat takes treating understanding as real work, not overhead. You read and challenge the codebase instead of just stacking features on it. You take nothing you did not write for granted. You watch the competition and what’s coming next, and you bring it in and ship it.\n\nUse AI agents if they help. They are productivity tools, not code owners. You are."},{"id":"e260b061-68b2-43d5-9f11-45261234f901","title":"Mechanical Engineer","department":"Engineering & Research","team":"Robotics","employmentType":"FullTime","location":"Bay Area","secondaryLocations":[],"publishedAt":"2026-04-30T18:40:29.371+00:00","isListed":true,"isRemote":false,"workplaceType":"OnSite","address":{"postalAddress":{"addressRegion":"California","addressCountry":"United States","addressLocality":"San Carlos"}},"jobUrl":"https://jobs.ashbyhq.com/genesis/e260b061-68b2-43d5-9f11-45261234f901","applyUrl":"https://jobs.ashbyhq.com/genesis/e260b061-68b2-43d5-9f11-45261234f901/application","descriptionHtml":"What You’ll Do
Design, prototype, and build robotic hardware systems optimized for performance and reliability
Develop scalable data collection platforms with robotics engineers to accelerate model learning
Improve simulation fidelity through system identification and close collaboration with the simulation team
Maintain and evolve in-house hardware infrastructure for maximum uptime and throughput
Collaborate with a team of driven individuals committed to building general-purpose Physical AI
What You’ll Bring
Strong engineering fundamentals and first-principles problem solving
Exceptional ownership and initiative—finding and solving problems independently
Broad experience across diverse hardware form factors and platforms
Systems-level mindset with a focus on robustness, scalability, and seamless integration
Track record of shipping fast without cutting corners on quality
Bonus: hands-on experience with actuators, PCB design, embedded systems, or electromechanical integration
Drive down wall-clock time to convergence by profiling and eliminating bottlenecks across the foundation model training stack stack, from data pipelines to GPU kernels
Design, build, and optimize distributed training systems (PyTorch) for multi-node GPU clusters, ensuring scalability, robustness, and high utilization
Implement efficient low-level code (CUDA, cuDNN, Triton, custom kernels) and integrate it seamlessly into high-level training frameworks
Optimize workloads for hardware efficiency: CPU/GPU compute balance, memory management, data throughput, and networking
Develop monitoring and debugging tools for large-scale runs, enabling rapid diagnosis of performance regressions and failures
Deep experience in distributed systems, ML infrastructure, or high-performance computing (8+ years)
Production-grade expertise in Python
Low-level performance mastery: CUDA/cuDNN/Triton, CPU–GPU interactions, data movement, and kernel optimization
Scaling at the frontier: experience with PyTorch and training jobs using data, context, pipeline, and model parallelism
System-level mindset with a track record of tuning hardware–software interactions for maximum utilization
Design, implement, and optimize the embedded control stack for general-purpose robots
Design motion planning and trajectory optimization algorithms for dynamic locomotion and manipulation
Build real-time state estimation pipelines for pose, contact, and force sensing, fusing heterogeneous sensor data under noise and uncertainty
Formulate and solve optimal control problems (nonlinear MPC, convex optimization, trajectory optimization) for high-performance and stable behavior
Build modular and robust software frameworks enabling rapid iteration between simulation and hardware
Lead debugging, tuning, and validation of controllers directly on physical robots
Passion for your craft and demonstrated excellence in control systems engineering
Extensive experience in designing, implementing, and deploying advanced control algorithms on real robotic system products (8+ years)
Deep expertise in dynamics, kinematics, optimal control, and state estimation
Strong command of hardware interfaces, sensors (IMUs, F/T sensors), and actuation technologies (motors, gearboxes, drivers)
Production-level mastery of C++ with a track record of building reliable, safety-critical software
Proven ability to bridge across hardware, software, and algorithms to deliver robust end-to-end systems
An open mind about the power of deep learning, reinforcement learning, and vision-language-action models as critical for general-purpose robotics
Bonus: Experience shipping humanoid robots or whole-body control systems
Bonus: Impactful published work in control theory, state estimation, or mathematical optimization
Bonus: Familiarity with parallel computation on GPUs to accelerate optimization
Build low-latency inference pipelines for on-device deployment, enabling real-time next-token and diffusion-based control loops in robotics
Design and optimize distributed inference systems on GPU clusters, pushing throughput with large-batch serving and efficient resource utilization
Implement efficient low-level code (CUDA, Triton, custom kernels) and integrate it seamlessly into high-level frameworks
Optimize workloads for both throughput (batching, scheduling, quantization) and latency (caching, memory management, graph compilation)
Develop monitoring and debugging tools to guarantee reliability, determinism, and rapid diagnosis of regressions across both stacks
Deep experience in distributed systems, ML infrastructure, or high-performance serving (8+ years)
Production-grade expertise in Python, with strong background in systems languages (C++/Rust/Go)
Low-level performance mastery: CUDA, Triton, kernel optimization, quantization, memory and compute scheduling
Proven track record scaling inference workloads in both throughput-oriented cluster environments and latency-critical on-device deployments
System-level mindset with a history of tuning hardware–software interactions for maximum efficiency, throughput, and responsiveness
About Genesis
We are a full-stack general-purpose robot company, born across San Francisco and Paris and bound by a single mission: to make general-purpose robots a reality, unlocking infinite physical labor for the world and freeing people for creativity, exploration, and the pursuits they love for their own sake.
We are French and American by birth and global by nature. We are backed by partners who share our vision, including Eclipse, Khosla Ventures, Bpifrance, and HSG, alongside Eric Schmidt, Xavier Niel, Daniela Rus, and Vladlen Koltun. Our people work across the Bay Area and Europe, building the engine that will teach robots to do real work in the real world. We recently unveiled Eno, our wheeled, dexterous robot powered by our GENE foundation model, and we are starting to put it to work with real customers.
The role
Everything our robots learn from passes through a station, a kit and a set of props, and all of it has to be somewhere, counted, working and ready when the floor needs it. Wearable collection kits (head-mounted cameras, gloves, wireless hubs) arrive by the hundred and go out to our floor, to partner sites and with every robot we deploy. Thousands of props and objects move between stations every week, and a task cannot run if its objects cannot be found. Spares, consumables, tools and fixtures keep the floor going. You own all of it: what comes in, what goes out, where it is, and what is ready.
This is the person who turns a pile of boxes into a working floor. You run receiving and shipping, keep the inventory true, build and serialize kits, run the props library, kit the stations, and make sure every new site opens with everything it needs on day one. As our in-house robot production line starts up, the role grows into maintaining its inventory as well: parts, kits and sub-assemblies for building robots, not only for collecting data.
What you'll do
Receiving and shipping: incoming deliveries checked against the order, serial numbers captured, items put away and recorded the same day; outbound shipments to partner sites, customers, suppliers and our Paris office packed, documented and tracked, with the paperwork international shipments need.
Kit building and serialization: assemble collection kits from components, test that each one powers on and pairs, serialize and label them, and keep the record of which kit is where and in what state.
Inventory: kits, devices, spares, consumables, tools and fixtures in one system with locations, counts and minimums; cycle counts on a schedule; reorder before the floor runs out; returns and repairs to suppliers tracked to closure.
Props and objects library: every object used in a task with an ID, a home location and a photo; checked out to stations and checked back in; new objects sourced, received and added when a task needs them; the library kept in sync with the task list.
Station kitting: each station type has a bill of materials; you kit new stations from it, keep running stations stocked, and support moves and re-layouts.
New sites and deployments: build the kit packages for partner sites and robot deployments to the checklist, ship them, and confirm receipt and setup with the site.
Production inventory, as the line starts: receive and store production parts and sub-assemblies, kit them to the build schedule, and keep the production stockroom accurate.
Records and systems: asset tags, barcodes or QR on everything that moves, the inventory system kept current, a weekly count of what is in stock, in use, in repair and in transit.
Space: receiving, inventory, stockroom and staging areas organized and safe, with a place for everything.
What we're looking for
Three or more years in shipping, receiving, inventory or kitting in a hardware, lab, manufacturing or field-operations environment.
You have run an inventory system, done cycle counts, and kept counts accurate under a fast pace; barcode or QR workflows are familiar.
You have built kits to a bill of materials and shipped them to deadlines, including internationally, and you know what paperwork a customs broker will ask for.
You are organized to the point that other people rely on it, and you notice when a count is off.
You are comfortable with electronics and small devices: handling, basic power-on testing, labeling, packing them so they arrive working.
You work well with operators, technicians and engineers, and you communicate clearly.
Bonus: ERP or inventory software setup; forklift or pallet jack certification; experience supporting a robotics, R&D or production floor.
Working conditions
Receiving, inventory and production environments; prolonged standing; lifting and moving boxes and equipment; occasional travel to partner sites.
Why this role
You will build the inventory and kitting operation for a floor that is growing every month and for the production line that follows it, and the kits you build go to every site and every robot we deploy.
How we work
We are always looking for people who are driven to work on hard problems and want to shape what the world will look like in ten years. A few of the values that guide us:
Limitless ambition. We hold ourselves to the highest bar in everything we build.
Relentless urgency. Iterate faster, learn faster. Time is the denominator.
Type-2 Fun. Take the work seriously, not yourself. Embrace the grind and find joy in the struggle.
Radical transparency. Candid communication builds trust, and direct feedback drives growth.
Multiculturalism. We come from across the world. Be curious. Meet people on their terms.
Team over ego. Help others, share context, and win as a team.
A joyful journey. We take pride in the craft, enjoy the ride, and make space to laugh. Happiness is part of how we get there.
The role is part-time and contractor-based (1–2 days a week), with the option to ramp up. Minimum duration is 3 months, renewable. A great fit for a student.
Role Overview
Genesis is a global physical AI lab and full-stack robotics company. We build generalist robots to unlock unlimited physical labor, so humans can focus on creativity, curiosity, and what we love.
As a Robot Teleoperation Specialist, your primary responsibility is to operate robots during data collection sessions using teleoperation equipment. You’ll execute predefined tasks by following standard operating procedures, monitor robot behavior during operation, and ensure sessions are run safely and consistently.
This is a highly hands-on, execution-focused role. Technical setup, hardware maintenance, and data infrastructure are handled by other teams — when you arrive, everything is ready for you to teleoperate.
This role is onsite in Paris, FR.
Responsibilities
Operate robots using teleoperation equipment according to predefined tasks and SOPs
Execute teleoperation sessions reliably and consistently to support data collection
Observe robot behavior during operation and flag unexpected or unsafe behavior
Follow all safety guidelines to maintain a safe environment for yourself and the robot
Provide basic session feedback or notes when required
Maintain focus and quality during repeated teleoperation tasks
Collaborate with operations staff and follow direction from session leads
Requirements
Strong attention to detail and ability to follow instructions precisely
Comfortable working on your feet and using teleoperation or VR-style equipment
High consistency and reliability in executing repetitive tasks
Ability to stay focused for extended teleoperation sessions
Clear written and verbal communication skills
Coachable, curious, and open to feedback
Adaptable to changing task requirements in an R&D environment
Bonus Qualifications
Prior experience with robot teleoperation, simulation, or VR systems
Strong hand-eye coordination and spatial awareness
Ability to learn new physical tasks quickly and improve execution over time
Design, build, and maintain foundational frameworks and tools to empower expert teams to experiment fast and turn ideas into full-fledged and production-ready systems.
Collaborate with expert teams to validate uses cases and build robust solutions, bringing cross-team perspective and aiming for modular and reusable components.
Identify and mitigate high-level code design flaws and development workflow inefficiencies that cause friction, hinder productivity, and create silos across the organization.
Advocate for good practices and maintain high code quality through code reviews, documentation, and training.
Excellent software engineering skills (e.g., in Python, Go, or Rust).
Extensive experience designing, building, and maintaining modular and reusable systems.
Experience collaborating with engineers and researchers of any fields, understanding complex requirements, and providing clear and simple solutions.
Ability to implement large-impact changes across the organization with minimal disruption.
Experience building internal tooling to accelerate AI research and development workflows (e.g., debugging, evaluation, experiment tracking and versioning, CI automation, etc.).
Design, implement, and optimize the embedded control stack for general-purpose robots
Design motion planning and trajectory optimization algorithms for dynamic locomotion and manipulation
Build real-time state estimation pipelines for pose, contact, and force sensing, fusing heterogeneous sensor data under noise and uncertainty
Formulate and solve optimal control problems (nonlinear MPC, convex optimization, trajectory optimization) for high-performance and stable behavior
Build modular and robust software frameworks enabling rapid iteration between simulation and hardware
Lead debugging, tuning, and validation of controllers directly on physical robots
Passion for your craft and demonstrated excellence in control systems engineering
Extensive experience in designing, implementing, and deploying advanced control algorithms on real robotic system products (8+ years)
Deep expertise in dynamics, kinematics, optimal control, and state estimation
Strong command of hardware interfaces, sensors (IMUs, F/T sensors), and actuation technologies (motors, gearboxes, drivers)
Production-level mastery of C++ with a track record of building reliable, safety-critical software
Proven ability to bridge across hardware, software, and algorithms to deliver robust end-to-end systems
An open mind about the power of deep learning, reinforcement learning, and vision-language-action models as critical for general-purpose robotics
Bonus: Experience shipping humanoid robots or whole-body control systems
Bonus: Impactful published work in control theory, state estimation, or mathematical optimization
Bonus: Familiarity with parallel computation on GPUs to accelerate optimization
What You’ll Do
Develop LM/VLM-powered agents that autonomously interface with code, GUI, web tools, and design software to drive generative simulation
Build agentic systems that synthesize high-quality, diverse physical data by orchestrating simulation, generative models, and interactive tools
Build data pipelines to curate, generate, and preprocess datasets used for training and evaluating language- and vision-language-based agents
Collaborate with simulation and infrastructure teams to co-develop APIs, GUIs, and services that enable seamless interaction with physics engines and simulation environments
Drive innovation by staying at the forefront of LLM agents, autonomous tool use, and generative simulation — and translating cutting-edge research into robust, production-ready systems
Contribute to the creation of a generative simulation stack that empowers downstream applications in robotics, embodied AI, and physical reasoning
What You’ll Bring
A bold and imaginative vision for a next-generation paradigm of physical data synthesis— combining simulation, generative models, and autonomous agents
Deep curiosity and strong technical ownership, with a track record of driving complex, open-ended projects from concept to implementation
Experience with (multimodal) large language models, generative AI tools, agentic software design, GUI automation, program synthesis
Passion for inventing creative solutions at the edge of AI, simulation, and physical reasoning, with an eagerness to build systems that generate high-fidelity, useful data for embodied intelligence
Bonus: Familiarity with interactive design tools (e.g., CAD software, game engines) or simulation environments (e.g., physics engines, robotics simulators)
Quadrants is an open-source translation layer that turns ordinary Python into optimized machine code at runtime, across CPU (Arm64, x86) and GPU (AMDGPU, CUDA, Metal, Vulkan) backends. CUDA is the primary target, with x86 the strong runner-up, especially for real-time simulation. The others drive adoption and support fast prototyping for research.
The ambition is explicit: build the best “Python compiler” in the world for physics and performance-critical numerical computing. We are fully committed to it, and we intend to beat leading alternatives like Nvidia Warp, Numba, or Triton.
It’s a standalone project with its own roadmap, built in the open. Genesis, our physics simulator, is its first and top-priority user, the source of its hardest workloads and its benchmarks. But Quadrants is general-purpose infrastructure for anyone writing performance-critical numerical Python, not an internal tool that happens to be public.
The bet: high-performance code should feel natural to write. Engineers write plain Python in a NumPy or PyTorch style, and Quadrants makes it fast.
You own Quadrants end to end: the API, the IR and lowering, the optimization passes, the numerical primitives, and the roadmap.
Our ultimate objective: engineers write for readability and maintainability, never for performance. They write the version of an algorithm they are most comfortable with, and Quadrants rebuilds it into kernels that run as if an expert had hand-tuned them. Performance is the compiler’s job, not theirs. Getting there is a frontier problem, part research and part engineering, and we intend to define the state of the art on it, relentlessly pushing the limit further.
It calls for two things at once: the low-level GPU and compiler skill to make a single kernel optimal, and the ability to keep the whole system coherent as it grows.
The core tradeoff is settled: usability comes first. When usability and performance collide, usability wins, unless the cost is severe. We will limit what the system supports to go fast (static shapes and fixed-size allocation baked into kernels, say), but never the convenience of the API itself. Experts who want maximum performance and full control over the generated code get lower-level escape hatches, down to inline assembly, C, or PTX.
API & authoring experience. The surface that physics and numerical-computing engineers write against, and, as an open-source library, the product’s face to the world. Idiomatic, natural to Python users, never a separate language bolted onto Python. Clean decorators, legible constraints, sharp errors, real escape hatches.
IR, lowering & backends. A stable IR that lowers to multiple backends, with room to grow and the door kept open for differentiable simulation. Today the computation graph is expressed through Quadrants’ own constructs. Recovering it automatically from ordinary Python is hard, and a long-term goal rather than today’s reality.
Optimization. Turn naive code into strong execution plans: vectorization, register and shared-memory usage, barriers and atomics, cooperative threading, dead-buffer and redundant-traffic elimination, dropping needless synchronization. Move low-level optimization off users and into the platform.
Primitives. Cholesky, eigendecomposition, matmul, reductions, solver kernels. Each owned with its correctness, benchmarks, and regression tracking.
Watching Genesis and the compiler together. Hunt for performance wins on both sides, in Genesis and in Quadrants itself. Propose API changes that cut duplication, improve readability, and reduce platform-specific special-casing. Less to maintain, more that simply works.
The project as a product. Benchmark real workloads, track performance over time, catch regressions early. Stay close to users (Genesis first), find friction and missing abstractions, drive adoption, set direction.
You reason about whole systems and have taste for where complexity should live. You measure before you optimize, and you reach for memory traffic and launch overhead before FLOPs. You have opinions about API design. You write tests and benchmarks as you go, not afterward.
The ground that judgment stands on:
GPU. What optimal looks like and why: access patterns, registers, shared memory, barriers, atomics, threading. CPU intuition a plus.
Python & API taste. Idiomatic Python. A feel for what’s natural to Python users. APIs that free users rather than constrain them.
Compilers. The path from authoring to kernels (IR, lowering, optimization), weighed against maintainability.
And how you work: proactive and largely self-directed. You fetch context across the company instead of waiting for it. You’re comfortable talking to the engineers who use what you build, and at ease designing in the open for users you’ll never meet. Ready to grow into leading a small team (a couple of people, near term).
People adopt Quadrants from outside Genesis. Engineers write clearer code with less boilerplate. Kernels keep getting faster. Performance is tracked, not guessed. The architecture gets easier to extend, not harder.
How you work is up to you. What matters is that your ownership grows over time: blind spots shrink, the parts you can vouch for expand, technical debt goes down, the core stays sound.
That takes treating understanding as real work, not overhead. You read and challenge the codebase instead of just stacking features on it. You take nothing you did not write for granted. You watch the competition and what’s coming next, and you bring it in and ship it.
Use AI agents if they help. They are productivity tools, not code owners. You are.
","descriptionPlain":"WHAT QUADRANTS IS\n\nQuadrants is an open-source translation layer that turns ordinary Python into optimized machine code at runtime, across CPU (Arm64, x86) and GPU (AMDGPU, CUDA, Metal, Vulkan) backends. CUDA is the primary target, with x86 the strong runner-up, especially for real-time simulation. The others drive adoption and support fast prototyping for research.\n\nThe ambition is explicit: build the best “Python compiler” in the world for physics and performance-critical numerical computing. We are fully committed to it, and we intend to beat leading alternatives like Nvidia Warp, Numba, or Triton.\n\nIt’s a standalone project with its own roadmap, built in the open. Genesis, our physics simulator, is its first and top-priority user, the source of its hardest workloads and its benchmarks. But Quadrants is general-purpose infrastructure for anyone writing performance-critical numerical Python, not an internal tool that happens to be public.\n\nThe bet: high-performance code should feel natural to write. Engineers write plain Python in a NumPy or PyTorch style, and Quadrants makes it fast.\n\n \n\n\nTHE ROLE\n\nYou own Quadrants end to end: the API, the IR and lowering, the optimization passes, the numerical primitives, and the roadmap.\n\nOur ultimate objective: engineers write for readability and maintainability, never for performance. They write the version of an algorithm they are most comfortable with, and Quadrants rebuilds it into kernels that run as if an expert had hand-tuned them. Performance is the compiler’s job, not theirs. Getting there is a frontier problem, part research and part engineering, and we intend to define the state of the art on it, relentlessly pushing the limit further.\n\nIt calls for two things at once: the low-level GPU and compiler skill to make a single kernel optimal, and the ability to keep the whole system coherent as it grows.\n\nThe core tradeoff is settled: usability comes first. When usability and performance collide, usability wins, unless the cost is severe. We will limit what the system supports to go fast (static shapes and fixed-size allocation baked into kernels, say), but never the convenience of the API itself. Experts who want maximum performance and full control over the generated code get lower-level escape hatches, down to inline assembly, C, or PTX.\n\n \n\n\nWHAT YOU’LL OWN\n\n - API & authoring experience. The surface that physics and numerical-computing engineers write against, and, as an open-source library, the product’s face to the world. Idiomatic, natural to Python users, never a separate language bolted onto Python. Clean decorators, legible constraints, sharp errors, real escape hatches.\n\n - IR, lowering & backends. A stable IR that lowers to multiple backends, with room to grow and the door kept open for differentiable simulation. Today the computation graph is expressed through Quadrants’ own constructs. Recovering it automatically from ordinary Python is hard, and a long-term goal rather than today’s reality.\n\n - Optimization. Turn naive code into strong execution plans: vectorization, register and shared-memory usage, barriers and atomics, cooperative threading, dead-buffer and redundant-traffic elimination, dropping needless synchronization. Move low-level optimization off users and into the platform.\n\n - Primitives. Cholesky, eigendecomposition, matmul, reductions, solver kernels. Each owned with its correctness, benchmarks, and regression tracking.\n\n - Watching Genesis and the compiler together. Hunt for performance wins on both sides, in Genesis and in Quadrants itself. Propose API changes that cut duplication, improve readability, and reduce platform-specific special-casing. Less to maintain, more that simply works.\n\n - The project as a product. Benchmark real workloads, track performance over time, catch regressions early. Stay close to users (Genesis first), find friction and missing abstractions, drive adoption, set direction.\n\n\nWHO YOU ARE\n\nYou reason about whole systems and have taste for where complexity should live. You measure before you optimize, and you reach for memory traffic and launch overhead before FLOPs. You have opinions about API design. You write tests and benchmarks as you go, not afterward.\n\nThe ground that judgment stands on:\n\n - GPU. What optimal looks like and why: access patterns, registers, shared memory, barriers, atomics, threading. CPU intuition a plus.\n\n - Python & API taste. Idiomatic Python. A feel for what’s natural to Python users. APIs that free users rather than constrain them.\n\n - Compilers. The path from authoring to kernels (IR, lowering, optimization), weighed against maintainability.\n\nAnd how you work: proactive and largely self-directed. You fetch context across the company instead of waiting for it. You’re comfortable talking to the engineers who use what you build, and at ease designing in the open for users you’ll never meet. Ready to grow into leading a small team (a couple of people, near term).\n\n \n\n\nWHAT GREAT LOOKS LIKE\n\nPeople adopt Quadrants from outside Genesis. Engineers write clearer code with less boilerplate. Kernels keep getting faster. Performance is tracked, not guessed. The architecture gets easier to extend, not harder.\n\n \n\n\nWHAT WE’RE UNCOMPROMISING ABOUT\n\nHow you work is up to you. What matters is that your ownership grows over time: blind spots shrink, the parts you can vouch for expand, technical debt goes down, the core stays sound.\n\nThat takes treating understanding as real work, not overhead. You read and challenge the codebase instead of just stacking features on it. You take nothing you did not write for granted. You watch the competition and what’s coming next, and you bring it in and ship it.\n\nUse AI agents if they help. They are productivity tools, not code owners. You are."},{"id":"9719ee70-1c41-443f-858b-dba2fa690553","title":"Data Infrastructure","department":"Engineering & Research","team":"Data","employmentType":"FullTime","location":"Paris","secondaryLocations":[{"location":"London","address":{"postalAddress":{"addressCountry":"United Kingdom","addressLocality":"London"}}}],"publishedAt":"2026-04-30T18:37:05.871+00:00","isListed":true,"isRemote":true,"workplaceType":"Hybrid","address":{"postalAddress":{"addressRegion":"Paris","addressCountry":"France","addressLocality":"Paris"}},"jobUrl":"https://jobs.ashbyhq.com/genesis/9719ee70-1c41-443f-858b-dba2fa690553","applyUrl":"https://jobs.ashbyhq.com/genesis/9719ee70-1c41-443f-858b-dba2fa690553/application","descriptionHtml":"What You’ll Do
Design, build, and maintain large-scale data pipelines (batch and streaming) for robotics foundation model training and evaluation at petabyte scale
Own core data infrastructure: data model, storage systems, ingestion pipelines, transformation frameworks, and orchestration layers
Standardize data models and unify processing pipelines across real-world teleoperation and synthetic simulation datasets
Collaborate with a team of driven individuals committed to building general-purpose Physical AI
What You’ll Bring
Excellent software engineering skills (Python, Go, or similar)
Extensive experience designing, building, and maintaining large-scale data pipelines (8+ years)
Deep understanding of distributed systems (Spark, Kafka, or similar)
Extensive experience with data storage technologies (data lakes, warehouses, object stores like S3)
Experience running and maintaining production-grade infrastructure (Kubernetes, Terraform)
Bonus: Experience supporting AI systems, in particular embodied AI like self-driving
What You’ll Do
Develop and optimize a learning-based robotic manipulation control stack
Design and maintain a teleoperation system with smooth, precise motion and low latency
Train robotic policies for manipulation and locomotion with reinforcement learning and imitation learning
Deploy robotic policies and diagnose latency or bottlenecks in the control pipeline
Analyze and minimize the sim-to-real gap by co-optimizing simulation and real-world robot behavior
Collaborate with a team of driven individuals committed to building general-purpose Physical AI
What You’ll Bring
Passion for your craft and demonstrated excellence in robotics engineering
Exceptional ownership and initiative—finding and solving problems independently
Focus, attention to detail, patience, and a methodical approach to complex tasks
Production-level expertise in modern Python or C++
Extensive experience building and deploying real-world learning-based robotic systems (5+ years)
Bonus: Hands-on experience troubleshooting and maintaining robotic systems across mechanical, electrical, and software components
Drive down wall-clock time to convergence by profiling and eliminating bottlenecks across the foundation model training stack stack, from data pipelines to GPU kernels
Design, build, and optimize distributed training systems (PyTorch) for multi-node GPU clusters, ensuring scalability, robustness, and high utilization
Implement efficient low-level code (CUDA, cuDNN, Triton, custom kernels) and integrate it seamlessly into high-level training frameworks
Optimize workloads for hardware efficiency: CPU/GPU compute balance, memory management, data throughput, and networking
Develop monitoring and debugging tools for large-scale runs, enabling rapid diagnosis of performance regressions and failures
Deep experience in distributed systems, ML infrastructure, or high-performance computing (8+ years)
Production-grade expertise in Python
Low-level performance mastery: CUDA/cuDNN/Triton, CPU–GPU interactions, data movement, and kernel optimization
Scaling at the frontier: experience with PyTorch and training jobs using data, context, pipeline, and model parallelism
System-level mindset with a track record of tuning hardware–software interactions for maximum utilization
Build core simulation-centered infrastructure for robotics model development, evaluation, and iteration.
Design scalable systems for creating, running, and managing diverse simulation workloads across tasks, environments, and robots.
Develop reusable environment frameworks, task interfaces, and tooling that make simulation workflows more reliable and observable.
Work with the policy training team to bridge the sim-to-real gap
Develop a real-to-sim pipeline in collaboration with the asset generation team
Collaborate with a team of driven individuals committed to building general-purpose Physical AI
Passion for your craft and demonstrated excellence in physics simulation engineering
Exceptional ownership and initiative: finding and solving problems independently
Extensive experience in physics simulation, with a preference for expertise in rigid body simulation and FEM
Insights and requests from a robotic researcher/engineer’s perspective for improving user experience
Proficiency in CUDA and GPU programming
Built closed-loop or log-replay evaluation at scale.
Experience with robotics simulation tools such as genesis-world, Isaac Sim, MuJoCo, Newton, Drake, Gazebo, or similar platforms.
Experience with synthetic data, scene generation, asset pipelines, sim-to-real, or robotics benchmarks.
Experience with robot manipulation, mobile manipulation, humanoids, or embodied AI systems.
What You'll Do
Build machine learning models for robotics control end-to-end: data curation, careful evaluation, model architecture, training/inference stacks, rigorous experiments
Collaborate with simulation and real-world robotics teams to curate high-quality, diverse, and large-scale datasets
Curate the world’s best Internet-scale datasets for embodied perception and first-person robot video generation
Design new generative simulation techniques to expand simulation data scale and diversity, training and evaluating generative models of 3D objects and environments, and language/code models to generate tasks and reward functions
Collaborate with a team of driven individuals committed to building general-purpose Physical AI
What You'll Bring
Passion for your craft and demonstrated excellence in full-stack foundation model research and engineering
Exceptional ownership and initiative—finding and solving problems independently
Extensive experience pioneering new machine learning ideas or refining existing methods, supported by first-author publications or impactful projects (5+ years)
A relentless commitment to data and code quality, rigorous evaluation, and meticulous attention to detail
Production-level expertise in modern Python
Bonus: Experience with Vision-Language-Action models for robotics or web agents, embodied perception, or video generation
Genesis AI is building Nyx, a renderer that delivers offline quality at realtime performance. We're looking for a principal-level rendering engineer to own performance across its pipeline. Nyx is a hybrid renderer, but not a classic game-engine hybrid. You'll need to master the profiling tools and have the technical depth to know how to apply them to a pipeline they've never seen, and to build your own instrumentation where they fall short.
Own frame-time budgets and drive GPU/CPU rendering performance across all target platforms
Profile, diagnose, and optimize the full rendering pipeline: geometry, shading, lighting, post
Adapt profiling methodology to a non-standard hybrid pipeline; build custom instrumentation and counters where existing tools can't see
Evolve hybrid raster/ray-traced rendering paths
Set the technical bar: review architecture, mentor senior engineers, kill bad ideas early
Work directly with hardware vendors on low-level GPU issues and our target hardware
10+ years in real-time rendering, with shipped AAA titles where you owned rendering performance
Mastery of GPU profiling tools: PIX, Razor GPU/CPU, RGP, Nsight, and reading raw GPU counters, with the judgment to know what each tool can and can't tell you on an unfamiliar pipeline
Deep knowledge of modern GPU architectures (AMD RDNA, wave occupancy, cache behavior, async compute)
Strong command of D3D12, Vulkan, and at least one console graphics API at a low level
Hands-on experience building or optimizing hybrid rendering pipelines
Expert-level C++ and HLSL/GLSL; comfortable in shader disassembly when it matters
Experience profiling or optimizing non-standard pipelines: neural rendering, splatting, offline/realtime hybrids
Experience with upscalers and frame generation (DLSS, FSR) integration and tuning
Contributions to engine-level rendering architecture in UE, proprietary engines, or both
Build low-latency inference pipelines for on-device deployment, enabling real-time next-token and diffusion-based control loops in robotics
Design and optimize distributed inference systems on GPU clusters, pushing throughput with large-batch serving and efficient resource utilization
Implement efficient low-level code (CUDA, Triton, custom kernels) and integrate it seamlessly into high-level frameworks
Optimize workloads for both throughput (batching, scheduling, quantization) and latency (caching, memory management, graph compilation)
Develop monitoring and debugging tools to guarantee reliability, determinism, and rapid diagnosis of regressions across both stacks
Deep experience in distributed systems, ML infrastructure, or high-performance serving (8+ years)
Production-grade expertise in Python, with strong background in systems languages (C++/Rust/Go)
Low-level performance mastery: CUDA, Triton, kernel optimization, quantization, memory and compute scheduling
Proven track record scaling inference workloads in both throughput-oriented cluster environments and latency-critical on-device deployments
System-level mindset with a history of tuning hardware–software interactions for maximum efficiency, throughput, and responsiveness
Design, build, and maintain foundational frameworks and tools to empower expert teams to experiment fast and turn ideas into full-fledged and production-ready systems.
Collaborate with expert teams to validate uses cases and build robust solutions, bringing cross-team perspective and aiming for modular and reusable components.
Identify and mitigate high-level code design flaws and development workflow inefficiencies that cause friction, hinder productivity, and create silos across the organization.
Advocate for good practices and maintain high code quality through code reviews, documentation, and training.
Excellent software engineering skills (e.g., in Python, Go, or Rust).
Extensive experience designing, building, and maintaining modular and reusable systems.
Experience collaborating with engineers and researchers of any fields, understanding complex requirements, and providing clear and simple solutions.
Ability to implement large-impact changes across the organization with minimal disruption.
Experience building internal tooling to accelerate AI research and development workflows (e.g., debugging, evaluation, experiment tracking and versioning, CI automation, etc.).
We are a full-stack general-purpose robot company, born across San Francisco and Paris and bound by a single mission: to make general-purpose robots a reality, unlocking infinite physical labor for the world and freeing people for creativity, exploration, and the pursuits they love for their own sake.
We are French and American by birth and global by nature. We are backed by partners who share our vision, including Eclipse, Khosla Ventures, Bpifrance, and HSG, alongside Eric Schmidt, Xavier Niel, Daniela Rus, and Vladlen Koltun. Our people work across the Bay Area and Europe, building the engine that will teach robots to do real work in the real world. We recently unveiled Eno, our wheeled, dexterous robot powered by our GENE foundation model, and we are starting to put it to work with real customers.
We build general-purpose robots, and next year we start building them in volume. You will be the manufacturing engineer who takes the robot from engineering builds to production. Our CA facility is where new products are introduced: the process is designed, the first stations are built and proven, and the technicians are trained there. Volume production runs with a manufacturing partner, on their floor or on ours with their people, and you own the process in both cases. You design how the robot gets built, you lead the engineers and technicians who prove it, and you own the quality gate that decides when a unit leaves.
This is hands-on work. You design the first station, write the first work instruction, and bring the partner's team up to our standard. What exists today: an engineering build process with our design team, a production partner, a bill of materials and a plan. What does not exist yet is the production process, and that is the job.
Design the production process: layout and cycle time, line balancing, station and fixture design, end-of-line test and calibration, material flow.
Capacity planning: stations, equipment, people and hours by build phase, from the first units to rate, and the plan to add capacity ahead of demand.
Specify, source and commission fixtures, tooling and test equipment: requirements, RFQ with Supply Chain, acceptance at the vendor and on the floor.
Own process documentation under revision control: work instructions with photos and video, PFMEA, control plans, inspection plans, SOPs; keep the manufacturing BOM aligned with the engineering BOM and run engineering changes through to the floor.
Run the transfer from engineering builds to production: pilot builds, DFM and DFA feedback with the authority to push back, build readiness reviews at each phase gate, release to volume.
Package the process so it can be copied: one documented set of stations, fixtures, test, instructions and training that a partner's floor can run to the same yield as ours.
Build and lead the team: manufacturing engineers, test engineers and technicians; hiring pace, training and certification, trainer capacity, shift structure as volume grows.
Own the quality gate: first-pass yield, defect capture, root cause (8D, SPC, DOE where it earns its keep), closed-loop feedback to engineering and suppliers; nothing ships that quality has not passed.
Partner with Supply Chain on clear-to-build so no unit starts without a complete kit, and with the manufacturing partner so their team runs our process to our standard.
Stand up the systems as we grow: traceability by serial, MES-lite at first, then the real thing; the weekly numbers (yield, hours per unit, cost per unit, downtime, schedule adherence) and the readout to leadership.
You have taken a complex electromechanical product from prototype through pilot to volume and designed the process yourself, not inherited it. Robots, vehicles, drones, medical devices, industrial equipment all count.
You have transferred a process to a contract manufacturer and held them to your standard.
Strong manufacturing engineering foundation: time studies, line balancing, capacity planning, fixture and tooling design, test strategy, PFMEA and control plans, lean and continuous improvement; CAD and GD&T fluency.
You have hired and trained technicians, written certification criteria, and know that trained people are the real constraint on throughput.
You have run a quality gate and held it under schedule pressure.
You are on the floor every day and rigorous in the data.
8 or more years in manufacturing engineering or production, with time in a mature organization so you know what good looks like at scale, and 3 or more years leading people.
Bonus: actuators, gearboxes or precision assemblies; PLM, MES and ERP implementation; product safety certification processes.
Production and lab environments; travel to partner sites, up to 25%.
You build the way our robots get built, from the first station to rate, and the process and the team that do it are yours.
We are always looking for people who are driven to work on hard problems and want to shape what the world will look like in ten years. A few of the values that guide us:
Limitless ambition. We hold ourselves to the highest bar in everything we build.
Relentless urgency. Iterate faster, learn faster. Time is the denominator.
Type-2 Fun. Take the work seriously, not yourself. Embrace the grind and find joy in the struggle.
Radical transparency. Candid communication builds trust, and direct feedback drives growth.
Multiculturalism. We come from across the world. Be curious. Meet people on their terms.
Team over ego. Help others, share context, and win as a team.
A joyful journey. We take pride in the craft.
Take on ambiguous reliability, scalability, and efficiency challenges and drive solutions across SRE and development teams.
Build and run large-scale, massively distributed, fault-tolerant systems that keep Genesis platform reliable and performant for our customers.
Optimize existing systems, build infrastructure, and eliminate toil through automation to continuously improve uptime and rate of change.
Cultivate a culture of reliability throughout the organization, guiding technical decisions that balance system health with fast-moving product priorities.
Ensure the long-term health, maintainability, and reliability of services through capacity planning, performance analysis, and proactive incident prevention.
Strong software engineering skills (e.g., in Python, Go, or similar) with extensive experience designing, analyzing, and troubleshooting distributed systems.
Deep expertise with cloud computing platforms (e.g., Kubernetes, Cloud Functions) and Non-Abstract Large Systems Design (NALSD).
Experience leading complex, large-scale technical projects and providing technical leadership across teams.
Ability to apply coding, algorithms, and complexity analysis to solve ambiguous problems at scale with minimal disruption.
A collaborative, intellectually curious mindset — comfortable working across a wide variety of backgrounds and bringing cross-team perspective to build robust, reusable solutions.
Translate research and engineering progress into clear external communication
Partner closely with founders, business teams and engineers to prepare milestone announcements, launch materials, and ecosystem updates
Build and maintain a coherent public narrative around Genesis’s
Design and execute our presence at major AI, robotics, autonomy, and deep tech conferences
Produce high-fidelity system demonstrations that accurately reflect real-world performance
Run lab visits, partner briefings, and technical press sessions
Create long-form technical explainers, talks, and written materials grounded in engineering accuracy
Support recruiting by clearly articulating the company’s mission, roadmap, and standards
Establish editorial and visual standards that reflect rigor and discipline
Build the marketing function from zero and hire selectively as the company scales
Operate hands-on without layers of support or predefined processes
Required: Experience in a hardware, robotics, autonomy, aerospace, or similarly technical environment
Required: 0→1 startup experience building something from scratch in a high-uncertainty setting
Strong ability to communicate complex systems clearly without oversimplification
Experience supporting major technical launches or public milestones
Comfort working directly with engineers and research teams
High ownership, autonomy, and bias for execution
Strong judgment around credibility, confidentiality, and technical precision
Ability to operate independently without structured playbooks
Ideally: Experience at an autonomy, aerospace, robotics, or frontier hardware company
Ideally: Experience communicating safety, reliability, and deployment readiness
Ideally: Experience running technical demonstrations or field events
What You’ll Do
Design, build, and scale security infrastructure from the ground up across our systems, networks, endpoints, and products
Own and evolve security architecture across endpoint security, network security, application security, product security, and data security
Develop and maintain detection, monitoring, and incident response capabilities across cloud and on-prem environments
Establish security controls, policies, and operational processes that enable the company to scale securely
Partner closely with engineering teams to integrate security throughout the software development lifecycle (SDLC)
Drive vulnerability management programs, including identification, prioritization, remediation, and validation
Lead risk assessments across infrastructure, applications, vendors, and internal systems
Build and improve security education and awareness programs across the organization
Collaborate with a team of driven individuals committed to building general-purpose Physical AI
Own IT Infrastructure in the interim
What You’ll Do
Design, build, and scale security infrastructure from the ground up across our systems, networks, endpoints, and products
Own and evolve security architecture across endpoint security, network security, application security, product security, and data security
Develop and maintain detection, monitoring, and incident response capabilities across cloud and on-prem environments
Establish security controls, policies, and operational processes that enable the company to scale securely
Partner closely with engineering teams to integrate security throughout the software development lifecycle (SDLC)
Drive vulnerability management programs, including identification, prioritization, remediation, and validation
Lead risk assessments across infrastructure, applications, vendors, and internal systems
Build and improve security education and awareness programs across the organization
Collaborate with a team of driven individuals committed to building general-purpose Physical AI
Own IT Infrastructure in the interim
What You’ll Bring
Excellent software engineering and infrastructure skills (Python, Go, or similar)
Extensive experience building and operating security infrastructure in fast-moving environments
Deep understanding of modern security domains including endpoint, network, application, product, and data security
Experience designing and implementing detection and response systems
Strong understanding of security controls, risk management, and vulnerability management best practices
Experience integrating security into engineering and infrastructure workflows across the SDLC
Hands-on experience operating production infrastructure and cloud environments
Ability to operate independently, prioritize effectively, and build systems from zero to scale
Experience leading or mentoring security teams, while remaining highly hands-on as an individual contributor
Bonus: Experience securing AI systems, robotics platforms, or large-scale distributed infrastructure
Excellent software engineering and infrastructure skills (Python, Go, or similar)
Extensive experience building and operating security infrastructure in fast-moving environments
Deep understanding of modern security domains including endpoint, network, application, product, and data security
Experience designing and implementing detection and response systems
Strong understanding of security controls, risk management, and vulnerability management best practices
Experience integrating security into engineering and infrastructure workflows across the SDLC
Hands-on experience operating production infrastructure and cloud environments
Ability to operate independently, prioritize effectively, and build systems from zero to scale
Experience leading or mentoring security teams, while remaining highly hands-on as an individual contributor
Bonus: Experience securing AI systems, robotics platforms, or large-scale distributed infrastructure
Genesis is a physical AI research lab and full-stack robotics company on a mission to unlock unlimited physical labor. We're building a universal robotics foundation model that unifies high-fidelity physics simulation, generative modeling, and large-scale real-world robot data.
To get there, we need the best Research Scientists, SWEs, and Robotics engineers and we're going head-to-head with the top AI labs to win them.
That's the job: making the best engineers turn down everyone else.
You won't be doing it alone, but you'll be early enough to shape how we hire for years to come: the playbook, the bar, the tooling, the team. If you want to build a recruiting function rather than inherit one, keep reading.
Closing elite technical talent. Full-cycle ownership of Research Scientist, SWE, and Robotics searches, from strategy through offer close, for candidates who are not on the market and have no reason to leave their current lab. Until you talk to them.
Being the leaders' recruiting brain. Partner with founders and engineering/research leaders as a real advisor: shape the search, calibrate the bar, push back when needed.
Sourcing where the best actually are. Papers, GitHub, conferences, communities, networks, not job-board spray. You go find them.
Running a process candidates rave about. Senior researchers judge a company by how it hires. Make ours sharp, fast, and human.
Building the machine. Stand up scalable sourcing, screening, and reporting in Ashby, and automate the parts that don't need a human so you spend your time on the parts that do.
You've closed senior, hard-to-win technical hires in engineering, AI/ML, or research, in a competitive market, full-360.
You can talk credibly with technical stakeholders and earn the trust of people who are skeptical of recruiters.
You think in systems and automation: you'd rather build a workflow (or an AI-assisted one) than do the same manual task twice.
You're a self-starter who brings structure to ambiguity and moves fast without dropping quality.
You read data well, including pipeline health, conversion, and where searches stall, and you adjust.
You've hired Research Scientists, robotics, or deep-tech engineers specifically.
You've recruited across UK, France, and US.
You've used AI tooling to genuinely speed up sourcing or screening, and have opinions about what works.
You've been early at a startup and built something from zero.
Competitive base salary plus meaningful equity
Visa sponsorship and relocation support
Hybrid: 3-4+ days per week in our office
A front-row seat to one of the hardest, most important problems in AI, and the chance to build the recruiting team behind it
Genesis is an equal opportunity employer. The robots we build will serve everyone, and we're building a team that reflects that.
","descriptionPlain":"ABOUT THE ROLE\n\nGenesis is a physical AI research lab and full-stack robotics company on a mission to unlock unlimited physical labor. We're building a universal robotics foundation model that unifies high-fidelity physics simulation, generative modeling, and large-scale real-world robot data. \n\nTo get there, we need the best Research Scientists, SWEs, and Robotics engineers and we're going head-to-head with the top AI labs to win them.\n\nThat's the job: making the best engineers turn down everyone else.\n\nYou won't be doing it alone, but you'll be early enough to shape how we hire for years to come: the playbook, the bar, the tooling, the team. If you want to build a recruiting function rather than inherit one, keep reading.\n\n\n\nWHAT YOU'LL OWN\n\n - Closing elite technical talent. Full-cycle ownership of Research Scientist, SWE, and Robotics searches, from strategy through offer close, for candidates who are not on the market and have no reason to leave their current lab. Until you talk to them.\n\n - Being the leaders' recruiting brain. Partner with founders and engineering/research leaders as a real advisor: shape the search, calibrate the bar, push back when needed.\n\n - Sourcing where the best actually are. Papers, GitHub, conferences, communities, networks, not job-board spray. You go find them.\n\n - Running a process candidates rave about. Senior researchers judge a company by how it hires. Make ours sharp, fast, and human.\n\n - Building the machine. Stand up scalable sourcing, screening, and reporting in Ashby, and automate the parts that don't need a human so you spend your time on the parts that do.\n \n\n\nYOU'LL THRIVE HERE IF\n\n - You've closed senior, hard-to-win technical hires in engineering, AI/ML, or research, in a competitive market, full-360.\n\n - You can talk credibly with technical stakeholders and earn the trust of people who are skeptical of recruiters.\n\n - You think in systems and automation: you'd rather build a workflow (or an AI-assisted one) than do the same manual task twice.\n\n - You're a self-starter who brings structure to ambiguity and moves fast without dropping quality.\n\n - You read data well, including pipeline health, conversion, and where searches stall, and you adjust.\n \n\n\nBONUS POINTS\n\n - You've hired Research Scientists, robotics, or deep-tech engineers specifically.\n\n - You've recruited across UK, France, and US.\n\n - You've used AI tooling to genuinely speed up sourcing or screening, and have opinions about what works.\n\n - You've been early at a startup and built something from zero.\n \n\n\nTHE OFFER\n\n - Competitive base salary plus meaningful equity\n\n - Visa sponsorship and relocation support\n\n - Hybrid: 3-4+ days per week in our office\n\n - A front-row seat to one of the hardest, most important problems in AI, and the chance to build the recruiting team behind it\n \n\nGenesis is an equal opportunity employer. The robots we build will serve everyone, and we're building a team that reflects that."},{"id":"949a83de-9438-4049-aa3f-121cee9d39e3","title":"QA Engineer: Nyx Renderer & Genesis World","department":"Engineering & Research","team":"Engineering & Research","employmentType":"FullTime","location":"Paris","secondaryLocations":[],"publishedAt":"2026-06-16T11:47:39.082+00:00","isListed":true,"isRemote":true,"workplaceType":"Hybrid","address":{"postalAddress":{"addressRegion":"Paris","addressCountry":"France","addressLocality":"Paris"}},"jobUrl":"https://jobs.ashbyhq.com/genesis/949a83de-9438-4049-aa3f-121cee9d39e3","applyUrl":"https://jobs.ashbyhq.com/genesis/949a83de-9438-4049-aa3f-121cee9d39e3/application","descriptionHtml":"About Genesis AI
Genesis AI builds the simulation platform powering the next generation of robotics and embodied AI. Our flagship product, Genesis World, is a unified multi-physics simulation engine paired with Nyx, our in-house GPU-accelerated, path-traced photorealistic renderer. Together they provide a platform for training robotics and embodied AI at scale; letting teams simulate complex physical worlds with high fidelity and speed, from a laptop to datacenter-grade GPUs, and close the gap between simulation and the real world.
About the Role
We're hiring a Senior QA Engineer to own quality across Nyx and the Genesis World engines. You'll split your time evenly between hands-on exploratory testing and enhancing our automated test infrastructure. Because both systems run on GPUs and underpin robotics training and scientific workloads, correctness goes beyond \"does it run\": it's about visual fidelity, numerical accuracy, determinism, and reproducibility across hardware.
We're especially looking for someone with strong knowledge of 3D environments, assets, and rendering, who can reason about what \"visually and physically correct\" actually means.
What You'll Do
Maintain and evolve our automated test suites for the Nyx renderer and the Genesis World physics engine.
Develop image-comparison / perceptual-diff tooling to catch rendering regressions, and validation harnesses to catch simulation drift.
Validate numerical correctness, determinism, and reproducibility across GPUs, drivers, and CUDA versions.
Perform exploratory and manual testing on 3D scenes, assets, and complex multi-physics scenarios that are hard to automate.
Set up quality gates in CI for prebuilt wheels across platforms (Linux, Windows, Python 3.10–3.13).
Triage, reproduce, and clearly document defects; partner with engineers on root cause.
What We're Looking For
5+ years in software QA, with strong experience in both test automation and exploratory testing.
Solid 3D environment and asset knowledge, with familiarity with rendering and simulation/physics.
Strong Python and C++ skills; comfortable reading performance-sensitive code.
Understanding of testing challenges in GPU and floating-point systems; non-determinism, tolerance-based assertions, hardware/driver variance.
Experience building test infrastructure in CI/CD and managing cross-platform builds.
Excellent debugging instincts and clear, reproducible bug reports.
Nice to Have
Experience with GPU programming, or ray/path tracing concepts.
Background in computer graphics (PBR, color management, perceptual image metrics).
Familiarity with physics or scientific simulation.
Robotics or sim-to-real experience.
Build sim-ready digital twin environments for real work assets, scenes, and tasks.
Build a pipeline that can generate simulation environment at scale.
Tune physics, rendering, geometry, robot settings to reduce sim-to-real gap.
Work closely with simulation engine, rendering, asset, AI, and robotics teams to improve realism, coverage, and usefulness of simulation.
Develop pipeline for iterating on simulation environment to boost model performance.
Strong software engineering skills in Python, C++, or similar language.
Experience building simulation systems, robotics infrastructure, evaluation platforms, or large-scale experimentation tools.
Strong understanding of software architecture, data pipelines, distributed systems, and production engineering practices.
Ability to work across robotics, simulation, ML, and infrastructure teams to turn ambiguous needs into robust systems.
Built closed-loop or log-replay evaluation at scale.
Experience with robotics simulation tools such as genesis-world, Isaac Sim, MuJoCo, Newton, Drake, Gazebo, or similar platforms.
Experience with synthetic data, scene generation, asset pipelines, sim-to-real, or robotics benchmarks.
Experience with robot manipulation, mobile manipulation, humanoids, or embodied AI systems.
Genesis is a physical AI research lab and full-stack robotics company on a mission to unlock unlimited physical labor. We're building a universal robotics foundation model that unifies high-fidelity physics simulation, generative modeling, and large-scale real-world robot data.
To get there, we need the best Research Scientists, SWEs, and Robotics engineers and we're going head-to-head with the top AI labs to win them.
That's the job: making the best engineers turn down everyone else.
You won't be doing it alone, but you'll be early enough to shape how we hire for years to come: the playbook, the bar, the tooling, the team. If you want to build a recruiting function rather than inherit one, keep reading.
Closing elite technical talent. Full-cycle ownership of Research Scientist, SWE, and Robotics searches, from strategy through offer close, for candidates who are not on the market and have no reason to leave their current lab. Until you talk to them.
Being the leaders' recruiting brain. Partner with founders and engineering/research leaders as a real advisor: shape the search, calibrate the bar, push back when needed.
Sourcing where the best actually are. Papers, GitHub, conferences, communities, networks, not job-board spray. You go find them.
Running a process candidates rave about. Senior researchers judge a company by how it hires. Make ours sharp, fast, and human.
Building the machine. Stand up scalable sourcing, screening, and reporting in Ashby, and automate the parts that don't need a human so you spend your time on the parts that do.
You've closed senior, hard-to-win technical hires in engineering, AI/ML, or research, in a competitive market, full-360.
You can talk credibly with technical stakeholders and earn the trust of people who are skeptical of recruiters.
You think in systems and automation: you'd rather build a workflow (or an AI-assisted one) than do the same manual task twice.
You're a self-starter who brings structure to ambiguity and moves fast without dropping quality.
You read data well, including pipeline health, conversion, and where searches stall, and you adjust.
You've hired Research Scientists, robotics, or deep-tech engineers specifically.
You've recruited across UK, France, and US.
You've used AI tooling to genuinely speed up sourcing or screening, and have opinions about what works.
You've been early at a startup and built something from zero.
Competitive base salary plus meaningful equity
Visa sponsorship and relocation support
Hybrid: 3-4+ days per week in our office
A front-row seat to one of the hardest, most important problems in AI, and the chance to build the recruiting team behind it
Genesis is an equal opportunity employer. The robots we build will serve everyone, and we're building a team that reflects that.
","descriptionPlain":"ABOUT THE ROLE\n\nGenesis is a physical AI research lab and full-stack robotics company on a mission to unlock unlimited physical labor. We're building a universal robotics foundation model that unifies high-fidelity physics simulation, generative modeling, and large-scale real-world robot data. \n\nTo get there, we need the best Research Scientists, SWEs, and Robotics engineers and we're going head-to-head with the top AI labs to win them.\n\nThat's the job: making the best engineers turn down everyone else.\n\nYou won't be doing it alone, but you'll be early enough to shape how we hire for years to come: the playbook, the bar, the tooling, the team. If you want to build a recruiting function rather than inherit one, keep reading.\n\n\n\nWHAT YOU'LL OWN\n\n - Closing elite technical talent. Full-cycle ownership of Research Scientist, SWE, and Robotics searches, from strategy through offer close, for candidates who are not on the market and have no reason to leave their current lab. Until you talk to them.\n\n - Being the leaders' recruiting brain. Partner with founders and engineering/research leaders as a real advisor: shape the search, calibrate the bar, push back when needed.\n\n - Sourcing where the best actually are. Papers, GitHub, conferences, communities, networks, not job-board spray. You go find them.\n\n - Running a process candidates rave about. Senior researchers judge a company by how it hires. Make ours sharp, fast, and human.\n\n - Building the machine. Stand up scalable sourcing, screening, and reporting in Ashby, and automate the parts that don't need a human so you spend your time on the parts that do.\n \n\n\nYOU'LL THRIVE HERE IF\n\n - You've closed senior, hard-to-win technical hires in engineering, AI/ML, or research, in a competitive market, full-360.\n\n - You can talk credibly with technical stakeholders and earn the trust of people who are skeptical of recruiters.\n\n - You think in systems and automation: you'd rather build a workflow (or an AI-assisted one) than do the same manual task twice.\n\n - You're a self-starter who brings structure to ambiguity and moves fast without dropping quality.\n\n - You read data well, including pipeline health, conversion, and where searches stall, and you adjust.\n \n\n\nBONUS POINTS\n\n - You've hired Research Scientists, robotics, or deep-tech engineers specifically.\n\n - You've recruited across UK, France, and US.\n\n - You've used AI tooling to genuinely speed up sourcing or screening, and have opinions about what works.\n\n - You've been early at a startup and built something from zero.\n \n\n\nTHE OFFER\n\n - Competitive base salary plus meaningful equity\n\n - Visa sponsorship and relocation support\n\n - Hybrid: 3-4+ days per week in our office\n\n - A front-row seat to one of the hardest, most important problems in AI, and the chance to build the recruiting team behind it\n \n\nGenesis is an equal opportunity employer. The robots we build will serve everyone, and we're building a team that reflects that."},{"id":"7c821c7f-3f0d-4afd-8eee-6c47430c9d8d","title":"Sourcer, AI Research & Engineering","department":"Talent & People","team":"Talent & People","employmentType":"FullTime","location":"Paris","secondaryLocations":[{"location":"Paris","address":{"postalAddress":{"addressRegion":"Paris","addressCountry":"France","addressLocality":"Paris"}}}],"publishedAt":"2026-06-18T16:03:02.651+00:00","isListed":true,"isRemote":true,"workplaceType":"Hybrid","address":{"postalAddress":{"addressRegion":"Paris","addressCountry":"France","addressLocality":"Paris"}},"jobUrl":"https://jobs.ashbyhq.com/genesis/7c821c7f-3f0d-4afd-8eee-6c47430c9d8d","applyUrl":"https://jobs.ashbyhq.com/genesis/7c821c7f-3f0d-4afd-8eee-6c47430c9d8d/application","descriptionHtml":"Genesis is a physical AI research lab and full-stack robotics company on a mission to unlock unlimited physical labor. We're building a universal robotics foundation model that unifies high-fidelity physics simulation, generative modeling, and large-scale real-world robot data.
To get there, we need the best Research Scientists, SWEs, and Robotics engineers, and we're going head-to-head with the top AI labs to win them. It all starts at the top of the funnel: finding the people who aren't looking, and getting them to look.
You're the front line of how Genesis wins talent. The quality of who we hire is capped by the quality of who you put in front of our team, and you'll be early enough to define how we source for years to come: the channels, the playbook, the tooling, the bar at the top of the funnel.
Finding the people no one else can. Source elite Research Scientist, SWE, and Robotics/infra candidates who aren't on the market and have no reason to leave their current lab. Until your message lands.
Sourcing where the best actually are. Papers, GitHub, conferences, citation graphs, communities, networks, not job-board spray. You go find them, and you know how to read who's actually exceptional.
Outreach people answer. Write personalized, credible outreach that earns replies from senior researchers and engineers who ignore recruiters. You're the first impression of Genesis; make it sharp.
Building pipeline that converts. Map markets, build target lists, and keep healthy, calibrated pipelines moving for every active search. Hand recruiters and hiring managers candidates worth their time.
Building the machine. Stand up scalable, well-organized sourcing in Ashby, and automate the parts that don't need a human (list building, enrichment, sequencing) so you spend your time on judgment and outreach.
You've sourced senior, hard-to-win technical talent (engineering, AI/ML, or research) in a competitive market, and have the response rates to show for it.
You can read a technical profile and tell signal from noise: whose papers matter, whose GitHub is real, who's actually exceptional.
You write outreach that sounds like a human who did their homework, not a template.
You think in systems and automation: you'd build an AI workflow than run the same manual search twice.
You're a self-starter who brings structure to ambiguity and moves fast without dropping quality.
You read data well, including response rates, pipeline health, and where searches stall, and you adjust.
You've sourced Research Scientists, robotics, or deep-tech engineers specifically.
You've sourced across UK and France.
You've used AI tooling to genuinely speed up sourcing, and have opinions about what works.
You've been early at a startup and built something from zero.
Competitive base salary plus meaningful equity
Visa sponsorship and relocation support
Hybrid: 3-4+ days per week in our office
A front-row seat to one of the hardest, most important problems in AI, and the chance to build the sourcing engine behind it
We are a full-stack general-purpose robot company, born across San Francisco and Paris and bound by a single mission: to make general-purpose robots a reality, unlocking infinite physical labor for the world and freeing people for creativity, exploration, and the pursuits they love for their own sake.
We are French and American by birth and global by nature. We are backed by partners who share our vision, including Eclipse, Khosla Ventures, Bpifrance, and HSG, alongside Eric Schmidt, Xavier Niel, Daniela Rus, and Vladlen Koltun. Our people work across the Bay Area and Europe, building the engine that will teach robots to do real work in the real world. We recently unveiled Eno, our wheeled, dexterous robot powered by our GENE foundation model, and we are starting to put it to work with real customers.
General-purpose robots learn from real, dexterous human work, the kind of data that does not exist on the internet. We capture it with a glove built to the exact geometry of our robot hand: a person wears it, does the work, and the robot learns to do it too. As a Senior Data Collection Operator, you are the person who creates that data, and the person who figures out how to create it well.
This is not a button-pushing job. You will wear our gloves and cameras, teleoperate our hardware, and, just as importantly, stand up new tasks from scratch: design the setup, the props, and the procedure, then run it to the quality bar. You will work shoulder to shoulder with our model engineering and research teams, turning their data requests into real demonstrations and feeding what you learn on the floor back to them. You will anchor the floor and set the bar the rest of the team grows toward. This is hands-on, high-craft work for an experienced operator who wants to be at the absolute front line of the mission.
Create the best data on the floor. Run tasks to the acceptance bar and produce the demonstrations the model learns the most from.
Set up tasks from scratch. Take a rough ask and design the task setup, props, environment, and SOP, then get it collection-ready fast.
Work with research and model engineering. Translate their data requests into demonstrations, and feed your observations from the floor back so the next request is sharper.
Set the standard. Anchor the floor and show newer operators what great looks like, in craft and in consistency.
Support reliability testing. Partner with the reliability team to run real-robot testing when they need the floor.
Own your quality. Review your own work, take direct feedback, and keep raising the bar.
You have done this at a robotics or embodied-AI operation. You have collected and teleoperated at a real quality bar, fast and precise. There is a small world of operators who do this work at the frontier. If you are one of them, we want to talk.
You build tasks, not just run them. You can take a vague request and stand up a clean, repeatable task from scratch.
Comfortable with research and engineering teams. You have worked alongside model or research teams and can turn their requests into data and their data into feedback.
Exceptional attention to detail. You do it right every time, and you notice the moment something is off.
Comfortable on your feet. You can stand and work with the hardware for a full shift, and lift up to 50 lbs to stage environments and support the gear.
Technically handy. Comfortable with Linux, Python scripting to automate, and hardware assembly and troubleshooting.
Bonus: reliability testing. You have partnered with a reliability or test team to run real-robot testing. A strong plus, not a requirement.
We are always looking for people who are driven to work on hard problems and want to shape what the world will look like in ten years. A few of the values that guide us:
Limitless ambition. We hold ourselves to the highest bar in everything we build.
Relentless urgency. Iterate faster, learn faster. Time is the denominator.
Type-2 Fun. Take the work seriously, not yourself. Embrace the grind and find joy in the struggle.
Radical transparency. Candid communication builds trust, and direct feedback drives growth.
Multiculturalism. We come from across the world. Be curious. Meet people on their terms.
Team over ego. Help others, share context, and win as a team.
A joyful journey. We take pride in the craft, enjoy the ride, and make space to laugh. Happiness is part of how we get there.
We are a full-stack general-purpose robot company, born across San Francisco and Paris and bound by a single mission: to make general-purpose robots a reality, unlocking infinite physical labor for the world and freeing people for creativity, exploration, and the pursuits they love for their own sake.
We are French and American by birth and global by nature. We are backed by partners who share our vision, including Eclipse, Khosla Ventures, Bpifrance, and HSG, alongside Eric Schmidt, Xavier Niel, Daniela Rus, and Vladlen Koltun. Our people work across the Bay Area and Europe, building the engine that will teach robots to do real work in the real world. We recently unveiled Eno, our wheeled, dexterous robot powered by our GENE foundation model, and we are starting to put it to work with real customers.
We have built a frontier model and put Eno in front of the world, fast. Behind that is a data engine: the machine that turns a raw human demonstration into data the model is measurably better for. This role owns that engine.
A worn glove and a camera produce a raw demonstration, not training data. You will build the pipeline and the annotation operation that turn raw demonstrations into clean, labeled, training-ready data, and make it scale with automation rather than headcount. You will own the datasets, what gets annotated, and the ontology, how it gets labeled, bring vision-language models to bear on trajectory labeling and language grounding, and close the loop so the engine keeps making the model better. This role serves the whole operation, our own floors and our partner-funded collection.
Run the data engine. Own the loop from raw trajectory and video to training-ready datasets, with validation steps that guarantee clean, correctly labeled data.
Own datasets and ontology. Decide what gets annotated and how, designing the ontology with the model team for its training implications.
Automate with models. Use vision-language models for automated trajectory annotation, language grounding, and data synthesis, so the pipeline scales without linear headcount, while holding the quality bar.
Run the annotation operation. Stand up and scale labeling, internal and vendor, against a clear quality bar and a delivery schedule the model team can plan around.
Close the loop. Turn real-robot eval failures into targeted collection and annotation jobs, and prove the new data improves the model.
Own the metrics. Track inter-annotator agreement, label error rate, and throughput per annotator-hour, and drive them the right way.
You have scaled an annotation or data pipeline at a serious operation. Four or more years in data or ML pipelines, including time leading the work. At a frontier AI lab or a top data operation, you have taken raw robot or embodied data to training-ready at volume and you know exactly where it breaks. The people who have done this are a small group. If you are one, we want to talk.
You can build, not just manage. Strong Python (Pandas, NumPy, PyTorch) and SQL. You write the automation that shrinks the pipeline.
ML literacy. You understand training versus test, precision and recall, and overfitting well enough to design an ontology that helps the model, not just labels data.
Hands-on technical leadership. You can run a labeling operation and stay a hands-on contributor at the same time.
Comfortable with ambiguity and speed. You move fast in a research-paced environment and bring order to it.
We are always looking for people who are driven to work on hard problems and want to shape what the world will look like in ten years. A few of the values that guide us:
Limitless ambition. We hold ourselves to the highest bar in everything we build.
Relentless urgency. Iterate faster, learn faster. Time is the denominator.
Type-2 Fun. Take the work seriously, not yourself. Embrace the grind and find joy in the struggle.
Radical transparency. Candid communication builds trust, and direct feedback drives growth.
Multiculturalism. We come from across the world. Be curious. Meet people on their terms.
Team over ego. Help others, share context, and win as a team.
A joyful journey. We take pride in the craft, enjoy the ride, and make space to laugh. Happiness is part of how we get there.
We are a full-stack general-purpose robot company, born across San Francisco and Paris and bound by a single mission: to make general-purpose robots a reality, unlocking infinite physical labor for the world and freeing people for creativity, exploration, and the pursuits they love for their own sake.
We are French and American by birth and global by nature. We are backed by partners who share our vision, including Eclipse, Khosla Ventures, Bpifrance, and HSG, alongside Eric Schmidt, Xavier Niel, Daniela Rus, and Vladlen Koltun. Our people work across the Bay Area and Europe, building the engine that will teach robots to do real work in the real world. We recently unveiled Eno, our wheeled, dexterous robot powered by our GENE foundation model, and we are starting to put it to work with real customers.
The hardest part of building general-purpose robots is not the model. It is the data: real, dexterous, physical-world manipulation that does not live on the internet. Our unlock is a data-collection glove built to the exact geometry of our robot hand, unobtrusive enough that a person can wear it through a normal workday, so ordinary work becomes training data that transfers to the robot without loss. Paired with our own hardware, simulation, and models in one tight loop, that is the most scalable manipulation data engine in the world.
This is the role that runs that engine on the ground. You will own the system that turns stations, workforce, and facilities into one reliable operation, end to end, from the operator at the station to the accepted hour that lands in training. You will grow it from a single floor into a global, multi-shift operation that holds a world-class quality bar as it scales. We are looking for someone who has run data collection at scale and is hungry to do it again at the frontier of physical AI.
Run the floor end to end. Own scheduling, operator utilization, station uptime, and the throughput that delivers accepted hours at the quality bar, every week.
Build the operator bench. Recruit, level, and develop operators and the senior operators who run shifts, until the team is deep enough for multiple shifts and multiple sites.
Hold one standard everywhere. Make every floor we run and every partner site collect to the same playbook, the same metrics, and the same bar.
Turn problems into process. Root-cause what slows the floor and ship the SOPs, layouts, and tooling that make the next site ramp faster than the last.
Partner on quality and annotation. Work hand in hand with the Quality and Process and Annotation functions so collected hours convert cleanly into training-ready data.
Stand up partner-site collection. Rehearse and run the mock-runs and first deployments that bring partner operators onto our gear.
Report the truth. Build the reporting that tells the company, honestly, what we collected, at what quality, and at what cost.
You have built data collection at scale, and it shows. You have run a teleoperation, glove, or human-data operation at a frontier robotics lab, an embodied-AI company, or a top data operation, and you know the difference between hours collected and hours that actually train a model. There is a short list of places this work happens. If you have done it, we want to talk.
Throughput is in your blood. Scheduling, utilization, yield, and the hundred details of a working floor are second nature.
Quality is never optional. You have run an operation where the bar was real and measured, and you held it without slowing the line.
You run through walls. Bias for action, deep ownership, and a refusal to wait for permission. You see what needs doing and make it happen.
You build teams that win. You hire well, coach honestly, and people do their best work for you.
You work across sites and cultures. You are comfortable partnering with our teams and sites in Asia and Europe, and operating across time zones.
We are always looking for people who are driven to work on hard problems and want to shape what the world will look like in ten years. A few of the values that guide us:
Limitless ambition. We hold ourselves to the highest bar in everything we build.
Relentless urgency. Iterate faster, learn faster. Time is the denominator.
Type-2 Fun. Take the work seriously, not yourself. Embrace the grind and find joy in the struggle.
Radical transparency. Candid communication builds trust, and direct feedback drives growth.
Multiculturalism. We come from across the world. Be curious. Meet people on their terms.
Team over ego. Help others, share context, and win as a team.
A joyful journey. We take pride in the craft, enjoy the ride, and make space to laugh. Happiness is part of how we get there.
About the Role
Genesis AI is building Genesis World, a unified multi-physics simulation engine paired with Nyx, our in-house path traced photorealistic renderer that powers the next generation of robotics and embodied AI. We are looking for a Senior Prop Artist to model, sculpt, and texture the physical objects and digital twins that populate our simulated worlds. Because these assets are used as training data for machines operating in the physical world, close enough is not enough, meaning scale, proportion, material response, and geometric accuracy must all hold up under scrutiny. This is a craft heavy role where you will apply game caliber art skills beyond entertainment to build beautiful, efficient assets that perform in real time and directly narrow the gap between simulation and reality.
What You'll Do
Create photoreal props for digital twins end-to-end: modeling, high-poly sculpting, retopology, UVs, and PBR texturing.
Recreate real-world objects from reference and photogrammetry, matching real dimensions, materials, and wear.
Clean up, retopologize, and optimize scanned and existing assets to meet real-time budgets without losing fidelity.
Author assets to strict poly, texture, and memory budgets suited to GPU-accelerated real-time rendering.
Partner with rendering and physics engineers to make sure assets are visually and physically correct in-engine.
Maintain consistency, naming, and quality standards across a growing library of props.
What We're Looking For
5+ years as a prop / hard-surface / environment artist, with shipped work built under AAA real-time engine constraints (games, not VFX).
Strong modeling skills in Maya and/or Blender and sculpting expertise in ZBrush.
Mastery of PBR texturing in Substance (Painter/Designer).
Photogrammetry and scan cleanup experience; comfortable turning raw scans into production-ready assets.
Deep understanding of real-time constraints: poly budgets, LODs, texture memory, and optimization tradeoffs.
A sharp eye for accuracy, scale, and material fidelity, and the discipline to hit them consistently.
Nice to Have
Familiarity with real-time / GPU-accelerated rendering pipelines and how assets behave in a path-traced renderer.
Interest or background in robotics, embodied AI, or sim-to-real.
We are a full-stack general-purpose robot company, born across San Francisco and Paris and bound by a single mission: to make general-purpose robots a reality, unlocking infinite physical labor for the world and freeing people for creativity, exploration, and the pursuits they love for their own sake.
We are French and American by birth and global by nature. We are backed by partners who share our vision, including Eclipse, Khosla Ventures, Bpifrance, and HSG, alongside Eric Schmidt, Xavier Niel, Daniela Rus, and Vladlen Koltun. Our people work across the Bay Area and Europe, building the engine that will teach robots to do real work in the real world. We recently unveiled Eno, our wheeled, dexterous robot powered by our GENE foundation model, and we are starting to put it to work with real customers.
Own the product roadmap for our core Physical AI capabilities, translating research breakthroughs into deployable, customer-facing products.
Sit at the intersection of research, robotics engineering, and field deployment to turn ambiguous technical tradeoffs into clear product decisions.
Define what \"ready for a customer site\" actually means by setting success metrics, reliability thresholds, and scope boundaries for each release.
Drive cross-functional alignment across distributed teams in research, engineering, and field operations.
Track record of shipping technical products where the primary challenge was managing complex tradeoffs, not filling out roadmap slides.
Deep enough technical understanding to read a policy evaluation, understand a sim-to-real gap, or participate in complex debug sessions and ask the right questions without needing to write the code yourself.
Strong bias toward the field over the whiteboard, preferring to evaluate a failed pick on a live customer floor over reading ten synthetic benchmark reports.
Exceptional written communication skills to write specs and decision logs that hold up when you're not in the room.
Comfort operating with incomplete information in a fast-moving, pre-paradigm field where answers aren't obvious.
Bonus: Prior experience in robotics, autonomy, or another domain where \"works in the lab\" and \"works in the world\" carry vastly different bars for success.
We are looking for a Technical Program Manager to work directly with our Head of Deployment to drive execution across our most critical customer deployments. You will operate at the center of complex engineering programs to coordinate AI, robotics, hardware, operations, and customer teams to move our systems from prototype to real-world production.
Partner directly with the Head of Deployment to drive end-to-end execution of customer deployments.
Own cross-functional programs spanning robotics software, AI, hardware, infrastructure, operations, and customer success.
Coordinate engineering milestones across multiple teams and ensure critical dependencies stay on track.
Drive site readiness, including integration planning, testing, validation, site logistics, and launch execution.
Engage directly with enterprise customers to translate technical requirements, report progress, and manage timelines.
Identify execution risks early and work across engineering teams to resolve technical and operational blockers.
Build scalable processes for deployment planning, release management, and customer onboarding.
Define how Genesis deploys and scales physical AI and robotics systems across diverse customer environments.
5+ years of experience in technical program management, systems engineering, or product operations for complex hardware/software platforms.
Technical depth to comfortably navigate technical discussions with software engineers, roboticists, and ML researchers.
Track record of leading cross-functional engineering programs from early planning through production rollout.
Clear written and verbal communication skills to align distributed engineering teams, leadership, and external enterprise stakeholders.
Strong systems thinking with a habit of bringing structure to ambiguous operational problems.
Willingness to travel to customer sites as needed to support critical deployment milestones.
Bonus: Direct experience deploying robotics or autonomous systems into real-world environments.
Bonus: Technical background in autonomy, perception, controls, simulation, machine learning, or field engineering.
Robots will learn in simulation before they hit the factory. Genesis-World is our bet on that future.
Genesis-World is an open-source, general-purpose simulation platform for physical AI from Genesis AI. One unified multi-physics engine: rigid bodies, FEM, MPM, particles, cloth, fluids. A robot arm can pour water onto sand, grasp a deformable object, or cut a soft body, all in the same simulation. Nyx, our in-house renderer, may be the most promising renderer for robotics out there: real-time photo-realistic rendering, advanced features like depth of field, and state-of-the-art techniques never seen before. Sensors of every kind: cameras, lidar, IMU, contact forces, temperature, plus arguably the most advanced tactile simulation available (paper). And the engine keeps growing: we are developing internally the most comprehensive and fastest Incremental Potential Contact (paper) solver for deformable body dynamics we know of, soon to be open-sourced. It powers real business applications, from full-fledged box packaging with labelling machine and all, to wire harnessing and lab automation, without any physics hack or compromise.
Everything is Python-first and runs anywhere. Kernels are written once, and Quadrants, our in-house JIT compiler, lowers them to CUDA, AMD ROCm, Apple Metal, Vulkan, x86, and ARM64. A single laptop or a datacenter. Massively batched GPU simulation for learning at scale, and complex non-batched scenes where CPU wins outright.
This is at the core of Genesis AI’s strategy. Evaluation is the bottleneck of scalable robotics: real hardware caps iteration at wall-clock time, but simulation turns it into a compute problem. Ours already runs two orders of magnitude faster than hardware (tens of thousands of episodes in half an hour instead of 200+ hours), while correlating with on-hardware rollouts at 89%. The north star: physical AI that improves at the speed of compute.
Simulation is still a hard sell in robotics. Outside a few success stories, like reinforcement learning for locomotion, most teams skip it, and friction is a big part of why: painful to use, painful to debug. We live that pain ourselves: debugging a failing experiment during policy evaluation takes 30 minutes, and there is no way to automatically generate a standalone reproduction script. That blocks adoption, even internally.
Your job is to kill that friction. You build the software engineering backbone of Genesis-World: everything that makes a large, fast-moving simulation codebase reliable, maintainable, frictionless. And it all ships in the open. Every hour of pain you remove, you remove for every roboticist who comes after you.
You join as core maintainer of Genesis-World, with shared stewardship of the whole platform from day one. Our physics team develops the algorithms, but just as importantly, you work to make everything around them excellent: infrastructure, tooling, APIs, architecture.
Make testing of a JIT-compiled engine scalable. Running our full suite is already taking hours because the ratio between compile time and runtime can be as bad as x10. Getting rid of this bottleneck is a serious challenge that will require a joint effort from the Genesis-World and the compiler teams.
Make any bug reproducible. Dump and reload simulations bit-exactly, across machines, across backends, even when a single timestep of a batched simulation can already max out VRAM. Turn any failing run into a standalone reproduction script, automatically.
Make autodiff first-class. Genesis-World is differentiable today, but partially: memory-hungry, slower than it could be, and hard to maintain due to manual operation recording. The project spans the engine and the compiler: you drive the requirements on Quadrants and iterate in a tight loop with the compiler team until the integration is solid.
Edit scenes at frozen time. See the scene, move entities and robots around, iterate on layout, all without building the scene or compiling the engine. Authoring today means a full build cycle. Make it instant.
Make the simulator extensible. First-class plugins for physics solvers and interactive viewer backends, including Nyx. Sensors already work this way. Generalize it, and Genesis-World becomes a platform others build on.
Give robots real actuation. Motors and MIMO transmissions, cleanly abstracted at the interface between users and the physics engine.
Make time a local variable. Today the clock and the timestep are global to the engine, which forbids whole classes of speedups: environments cannot step independently, the timestep cannot adapt to the state, and individual solvers cannot advance at their own rate. Free the clock down to the sub-scene level and let the simulation fly.
Make scenes come alive. Events triggered by physics: press the button and the microwave door pops open, squeeze the trigger and the drill spins.
Keep the core sharp. Refactoring legacy subsystems, stronger typing, consistent coding style, less Python overhead on hot paths, full scene serialization, telemetry and replay.
Beyond the code: you review issues and pull requests with the core team, answer questions from users and contributors, and can join the push on adoption.
You care about the engineer using what you build as much as the thing itself. A slow test, a cryptic error, a manual step: you see it, you go after the root cause, and you remove it for everyone. You have opinions about API design.
You are comfortable with physics and robotics: the right design choices take a working picture of what a simulator like Genesis-World does, how it works, and how it is used, a picture you can partly build on the job.
Strong Python engineering on large codebases: architecture, refactoring, typing, packaging, performance profiling (py-spy, Nsight, Xcode Instruments, RenderDoc).
Developer tooling, test infrastructure, or CI at scale.
Enough fluency with JIT compilation, code generation, or GPU computing stacks to reason about compilation caching, kernel dispatch, and startup cost. You will work closely with Quadrants.
At ease debugging across the full platform matrix: Windows, Linux, and macOS, on x86 and arm64, over CUDA, AMD, and Apple Metal (and possibly Intel XPU via Vulkan soon)
Open-source reflexes: triaging issues, reviewing external pull requests, communicating with a community.
Bonus points: experience with other robotics simulators (Drake, MuJoCo, Newton, Isaac, RaiSim, Jiminy, Brax). PyTorch, our user-facing interface. Other JIT compilers such as Triton, JAX, or Numba. 3D tooling and interactive editors (Maya, Blender, Houdini). Serialization formats (USD, glTF, MJCF/URDF). C++, CUDA, or Metal. Developer relations or community work.
","descriptionPlain":"WHAT WE’RE BUILDING\n\nRobots will learn in simulation before they hit the factory. Genesis-World is our bet on that future.\n\nGenesis-World https://github.com/Genesis-Embodied-AI/genesis-world is an open-source, general-purpose simulation platform for physical AI from Genesis AI https://www.genesis.ai/. One unified multi-physics engine: rigid bodies, FEM, MPM, particles, cloth, fluids. A robot arm can pour water onto sand, grasp a deformable object, or cut a soft body, all in the same simulation. Nyx, our in-house renderer, may be the most promising renderer for robotics out there: real-time photo-realistic rendering, advanced features like depth of field, and state-of-the-art techniques never seen before. Sensors of every kind: cameras, lidar, IMU, contact forces, temperature, plus arguably the most advanced tactile simulation available (paper https://openreview.net/forum?id=md8q2kfZHP). And the engine keeps growing: we are developing internally the most comprehensive and fastest Incremental Potential Contact (paper https://ipc-sim.github.io) solver for deformable body dynamics we know of, soon to be open-sourced. It powers real business applications, from full-fledged box packaging with labelling machine and all, to wire harnessing and lab automation, without any physics hack or compromise.\n\nEverything is Python-first and runs anywhere. Kernels are written once, and Quadrants, our in-house JIT compiler, lowers them to CUDA, AMD ROCm, Apple Metal, Vulkan, x86, and ARM64. A single laptop or a datacenter. Massively batched GPU simulation for learning at scale, and complex non-batched scenes where CPU wins outright.\n\nThis is at the core of Genesis AI’s strategy https://www.genesis.ai/blog/the-role-of-simulation-in-scalable-robotics-genesis-world-10-and-the-path-forward. Evaluation is the bottleneck of scalable robotics: real hardware caps iteration at wall-clock time, but simulation turns it into a compute problem. Ours already runs two orders of magnitude faster than hardware (tens of thousands of episodes in half an hour instead of 200+ hours), while correlating with on-hardware rollouts at 89%. The north star: physical AI that improves at the speed of compute.\n\n\n\n\nTHE ROLE\n\nSimulation is still a hard sell in robotics. Outside a few success stories, like reinforcement learning for locomotion, most teams skip it, and friction is a big part of why: painful to use, painful to debug. We live that pain ourselves: debugging a failing experiment during policy evaluation takes 30 minutes, and there is no way to automatically generate a standalone reproduction script. That blocks adoption, even internally.\n\nYour job is to kill that friction. You build the software engineering backbone of Genesis-World: everything that makes a large, fast-moving simulation codebase reliable, maintainable, frictionless. And it all ships in the open. Every hour of pain you remove, you remove for every roboticist who comes after you.\n\nYou join as core maintainer of Genesis-World, with shared stewardship of the whole platform from day one. Our physics team develops the algorithms, but just as importantly, you work to make everything around them excellent: infrastructure, tooling, APIs, architecture.\n\n\n\n\nTHE PROBLEMS WAITING FOR YOU\n\n - Make testing of a JIT-compiled engine scalable. Running our full suite is already taking hours because the ratio between compile time and runtime can be as bad as x10. Getting rid of this bottleneck is a serious challenge that will require a joint effort from the Genesis-World and the compiler teams.\n\n - Make any bug reproducible. Dump and reload simulations bit-exactly, across machines, across backends, even when a single timestep of a batched simulation can already max out VRAM. Turn any failing run into a standalone reproduction script, automatically.\n\n - Make autodiff first-class. Genesis-World is differentiable today, but partially: memory-hungry, slower than it could be, and hard to maintain due to manual operation recording. The project spans the engine and the compiler: you drive the requirements on Quadrants and iterate in a tight loop with the compiler team until the integration is solid.\n\n - Edit scenes at frozen time. See the scene, move entities and robots around, iterate on layout, all without building the scene or compiling the engine. Authoring today means a full build cycle. Make it instant.\n\n - Make the simulator extensible. First-class plugins for physics solvers and interactive viewer backends, including Nyx. Sensors already work this way. Generalize it, and Genesis-World becomes a platform others build on.\n\n - Give robots real actuation. Motors and MIMO transmissions, cleanly abstracted at the interface between users and the physics engine.\n\n - Make time a local variable. Today the clock and the timestep are global to the engine, which forbids whole classes of speedups: environments cannot step independently, the timestep cannot adapt to the state, and individual solvers cannot advance at their own rate. Free the clock down to the sub-scene level and let the simulation fly.\n\n - Make scenes come alive. Events triggered by physics: press the button and the microwave door pops open, squeeze the trigger and the drill spins.\n\n - Keep the core sharp. Refactoring legacy subsystems, stronger typing, consistent coding style, less Python overhead on hot paths, full scene serialization, telemetry and replay.\n\nBeyond the code: you review issues and pull requests with the core team, answer questions from users and contributors, and can join the push on adoption.\n\n\n\n\nWHO YOU ARE\n\nYou care about the engineer using what you build as much as the thing itself. A slow test, a cryptic error, a manual step: you see it, you go after the root cause, and you remove it for everyone. You have opinions about API design.\n\nYou are comfortable with physics and robotics: the right design choices take a working picture of what a simulator like Genesis-World does, how it works, and how it is used, a picture you can partly build on the job.\n\n - Strong Python engineering on large codebases: architecture, refactoring, typing, packaging, performance profiling (py-spy, Nsight, Xcode Instruments, RenderDoc).\n\n - Developer tooling, test infrastructure, or CI at scale.\n\n - Enough fluency with JIT compilation, code generation, or GPU computing stacks to reason about compilation caching, kernel dispatch, and startup cost. You will work closely with Quadrants.\n\n - At ease debugging across the full platform matrix: Windows, Linux, and macOS, on x86 and arm64, over CUDA, AMD, and Apple Metal (and possibly Intel XPU via Vulkan soon)\n\n - Open-source reflexes: triaging issues, reviewing external pull requests, communicating with a community.\n\nBonus points: experience with other robotics simulators (Drake, MuJoCo, Newton, Isaac, RaiSim, Jiminy, Brax). PyTorch, our user-facing interface. Other JIT compilers such as Triton, JAX, or Numba. 3D tooling and interactive editors (Maya, Blender, Houdini). Serialization formats (USD, glTF, MJCF/URDF). C++, CUDA, or Metal. Developer relations or community work."},{"id":"4ef65db2-2989-4167-a27e-2927f816eb4f","title":"Genesis-World: Core Physics Engineer","department":"Engineering & Research","team":"Simulation","employmentType":"FullTime","location":"Paris","secondaryLocations":[],"publishedAt":"2026-08-17T11:23:15.788+00:00","isListed":true,"isRemote":true,"workplaceType":"Hybrid","address":{"postalAddress":{"addressRegion":"Paris","addressCountry":"France","addressLocality":"Paris"}},"jobUrl":"https://jobs.ashbyhq.com/genesis/4ef65db2-2989-4167-a27e-2927f816eb4f","applyUrl":"https://jobs.ashbyhq.com/genesis/4ef65db2-2989-4167-a27e-2927f816eb4f/application","descriptionHtml":"Robots will learn in simulation before they hit the factory. Genesis-World is our bet on that future.
Genesis-World is an open-source, general-purpose simulation platform for physical AI from Genesis AI. One unified multi-physics engine: rigid bodies, FEM, MPM, particles, cloth, fluids. A robot arm can pour water onto sand, grasp a deformable object, or cut a soft body, all in the same simulation. Nyx, our in-house renderer, may be the most promising renderer for robotics out there: real-time photo-realistic rendering, advanced features like depth of field, and state-of-the-art techniques never seen before. Sensors of every kind: cameras, lidar, IMU, contact forces, temperature, plus arguably the most advanced tactile simulation available (paper). And the engine keeps growing: we are developing internally the most comprehensive and fastest Incremental Potential Contact (paper) solver for deformable body dynamics we know of, soon to be open-sourced. It powers real business applications, from full-fledged box packaging with labelling machine and all, to wire harnessing and lab automation, without any physics hack or compromise.
Everything is Python-first and runs anywhere. Kernels are written once, and Quadrants, our in-house JIT compiler, lowers them to CUDA, AMD ROCm, Apple Metal, Vulkan, x86, and ARM64. A single laptop or a datacenter. Massively batched GPU simulation for learning at scale, and complex non-batched scenes where CPU wins outright.
This is at the core of Genesis AI's strategy. Evaluation is the bottleneck of scalable robotics: real hardware caps iteration at wall-clock time, but simulation turns it into a compute problem. Ours already runs two orders of magnitude faster than hardware (tens of thousands of episodes in half an hour instead of 200+ hours), while correlating with on-hardware rollouts at 89%. The north star: physical AI that improves at the speed of compute.
You push the physics of Genesis-World forward. The mandate is clear: ship production-ready simulation capabilities that matter for the company's internal needs. Research applied end-to-end, from algorithm to merged, tested, documented code that real robot-learning pipelines depend on. Occasional groundbreaking research happens, notably through academic collaborations. But the core of the job is making the engine measurably better along five axes:
Speed. Algorithms that are not only faster but also smart enough to spend compute only where it matters across both time and space: larger stable timesteps, selective fidelity (adaptive across scales or simply hand-set), structure-aware solvers.
Completeness. No physics off limits: water, human animation, air flow, gravel, tendons, even body organs. Whatever the next use-case needs, the engine grows to cover it.
Fidelity. More realistic models: contact, friction, deformation, energy, actuation, materials…
Versatility. Extensible multi-physics without compromise on realism: all solvers in the scene coupled together at once, two-way and constraint-based. Write your own solver and it joins the scene like a native one, growing into an open solver ecosystem.
Scalability. From workstation to factory scale, and one day, city scale: thousands of interacting entities, batched across environments, without losing physical soundness.
Our ambition is to establish Genesis-World as the go-to simulator for physical AI, from companies and research labs to individuals.
Every fidelity for every physics. The same physics at every point of the speed-accuracy spectrum, from heavily batched training with XPBD or VBD to final validation with IPC. Same scene, same API, pick your tradeoff.
Invent physics level-of-detail (LOD). Rendering has had LOD for decades, physics is still waiting. Simulate at full fidelity what agents interact with and see, coarsely what they do not.
Heterogeneous environments. Every parallel world can hold a completely different model: different bodies, joints, and collision geometries.
Adaptive timesteps per island. Error-based control with Runge-Kutta Dopri5, and Time-of-Impact stepping during collision detection, as done in Jiminy.
Couple everything, exactly. Efficient and accurate two-way constraint-based coupling between heterogeneous grey-box solvers, using state-of-the-art methods like ADMM. Owning every solver in the stack is what makes it possible.
More scalable constraint solvers. Push rigid constraint solving beyond its current scalability ceiling (reference).
Unify contact resolution. Hydro-elastic compliance, unilateral constraints, and sequential impulses in the same framework, ideally under one generic formulation.
Closed kinematic loops without constraints. Handle loops intrinsically for numerical stability and speed, in the spirit of Kamino.
Day to day: you write your physics in plain Python and Quadrants makes it fast on every backend. And you validate it the hard way: analytical closed forms, other engines, real-world data.
You are a physicist and an engineer at once. You judge a method by whether it holds up in production at real scale, and you do not stop until it does. No blind spots: you relentlessly hunt down even the defect that looks insignificant, because it never is.
A strong background in physics-based simulation, preferably related to robotics: RBD, FEM, MPM, SPH, IPC, XPBD, VBD, ABD, plus constrained optimization and numerical integration of stiff systems.
A track record of shipping simulation code that others rely on, in an engine, in industry, or in a research codebase used beyond its authors.
Solid HPC programming (CPU and/or GPU), and an instinct for what makes a numerical method fast in practice, beyond complexity classes.
Rigor in validation: analytical closed forms, cross-engine consistency, real-world data.
Bonus points: publications in simulation, graphics, or robotics venues (SIGGRAPH, ICRA, IROS, CoRL, RSS). Contributions to an open-source physics engine.
","descriptionPlain":"WHAT WE'RE BUILDING\n\nRobots will learn in simulation before they hit the factory. Genesis-World is our bet on that future.\n\nGenesis-World https://github.com/Genesis-Embodied-AI/genesis-world is an open-source, general-purpose simulation platform for physical AI from Genesis AI https://www.genesis.ai/. One unified multi-physics engine: rigid bodies, FEM, MPM, particles, cloth, fluids. A robot arm can pour water onto sand, grasp a deformable object, or cut a soft body, all in the same simulation. Nyx, our in-house renderer, may be the most promising renderer for robotics out there: real-time photo-realistic rendering, advanced features like depth of field, and state-of-the-art techniques never seen before. Sensors of every kind: cameras, lidar, IMU, contact forces, temperature, plus arguably the most advanced tactile simulation available (paper https://openreview.net/forum?id=md8q2kfZHP). And the engine keeps growing: we are developing internally the most comprehensive and fastest Incremental Potential Contact (paper https://ipc-sim.github.io) solver for deformable body dynamics we know of, soon to be open-sourced. It powers real business applications, from full-fledged box packaging with labelling machine and all, to wire harnessing and lab automation, without any physics hack or compromise.\n\nEverything is Python-first and runs anywhere. Kernels are written once, and Quadrants, our in-house JIT compiler, lowers them to CUDA, AMD ROCm, Apple Metal, Vulkan, x86, and ARM64. A single laptop or a datacenter. Massively batched GPU simulation for learning at scale, and complex non-batched scenes where CPU wins outright.\n\nThis is at the core of Genesis AI's strategy https://www.genesis.ai/blog/the-role-of-simulation-in-scalable-robotics-genesis-world-10-and-the-path-forward. Evaluation is the bottleneck of scalable robotics: real hardware caps iteration at wall-clock time, but simulation turns it into a compute problem. Ours already runs two orders of magnitude faster than hardware (tens of thousands of episodes in half an hour instead of 200+ hours), while correlating with on-hardware rollouts at 89%. The north star: physical AI that improves at the speed of compute.\n\n\n\n\nTHE ROLE\n\nYou push the physics of Genesis-World forward. The mandate is clear: ship production-ready simulation capabilities that matter for the company's internal needs. Research applied end-to-end, from algorithm to merged, tested, documented code that real robot-learning pipelines depend on. Occasional groundbreaking research happens, notably through academic collaborations. But the core of the job is making the engine measurably better along five axes:\n\n - Speed. Algorithms that are not only faster but also smart enough to spend compute only where it matters across both time and space: larger stable timesteps, selective fidelity (adaptive across scales or simply hand-set), structure-aware solvers.\n\n - Completeness. No physics off limits: water, human animation, air flow, gravel, tendons, even body organs. Whatever the next use-case needs, the engine grows to cover it.\n\n - Fidelity. More realistic models: contact, friction, deformation, energy, actuation, materials…\n\n - Versatility. Extensible multi-physics without compromise on realism: all solvers in the scene coupled together at once, two-way and constraint-based. Write your own solver and it joins the scene like a native one, growing into an open solver ecosystem.\n\n - Scalability. From workstation to factory scale, and one day, city scale: thousands of interacting entities, batched across environments, without losing physical soundness.\n\nOur ambition is to establish Genesis-World as the go-to simulator for physical AI, from companies and research labs to individuals.\n\n\n\n\nTHE PROBLEMS WAITING FOR YOU\n\n - Every fidelity for every physics. The same physics at every point of the speed-accuracy spectrum, from heavily batched training with XPBD or VBD to final validation with IPC. Same scene, same API, pick your tradeoff.\n\n - Invent physics level-of-detail (LOD). Rendering has had LOD for decades, physics is still waiting. Simulate at full fidelity what agents interact with and see, coarsely what they do not.\n\n - Heterogeneous environments. Every parallel world can hold a completely different model: different bodies, joints, and collision geometries.\n\n - Adaptive timesteps per island. Error-based control with Runge-Kutta Dopri5, and Time-of-Impact stepping during collision detection, as done in Jiminy.\n\n - Couple everything, exactly. Efficient and accurate two-way constraint-based coupling between heterogeneous grey-box solvers, using state-of-the-art methods like ADMM. Owning every solver in the stack is what makes it possible.\n\n - More scalable constraint solvers. Push rigid constraint solving beyond its current scalability ceiling (reference https://www.roboticsproceedings.org/rss20/p108.pdf).\n\n - Unify contact resolution. Hydro-elastic compliance, unilateral constraints, and sequential impulses in the same framework, ideally under one generic formulation.\n\n - Closed kinematic loops without constraints. Handle loops intrinsically for numerical stability and speed, in the spirit of Kamino https://disneyresearch.github.io/kamino.\n\nDay to day: you write your physics in plain Python and Quadrants makes it fast on every backend. And you validate it the hard way: analytical closed forms, other engines, real-world data.\n\n\n\n\nWHO YOU ARE\n\nYou are a physicist and an engineer at once. You judge a method by whether it holds up in production at real scale, and you do not stop until it does. No blind spots: you relentlessly hunt down even the defect that looks insignificant, because it never is.\n\n - A strong background in physics-based simulation, preferably related to robotics: RBD, FEM, MPM, SPH, IPC, XPBD, VBD, ABD, plus constrained optimization and numerical integration of stiff systems.\n\n - A track record of shipping simulation code that others rely on, in an engine, in industry, or in a research codebase used beyond its authors.\n\n - Solid HPC programming (CPU and/or GPU), and an instinct for what makes a numerical method fast in practice, beyond complexity classes.\n\n - Rigor in validation: analytical closed forms, cross-engine consistency, real-world data.\n\nBonus points: publications in simulation, graphics, or robotics venues (SIGGRAPH, ICRA, IROS, CoRL, RSS). Contributions to an open-source physics engine."},{"id":"8ad72d05-e4ce-4cf8-b27b-818abf54c3ad","title":"Quadrants: Compiler Lead","department":"Engineering & Research","team":"Simulation","employmentType":"FullTime","location":"Paris","secondaryLocations":[],"publishedAt":"2026-08-17T11:24:44.467+00:00","isListed":true,"isRemote":true,"workplaceType":"Hybrid","address":{"postalAddress":{"addressRegion":"Paris","addressCountry":"France","addressLocality":"Paris"}},"jobUrl":"https://jobs.ashbyhq.com/genesis/8ad72d05-e4ce-4cf8-b27b-818abf54c3ad","applyUrl":"https://jobs.ashbyhq.com/genesis/8ad72d05-e4ce-4cf8-b27b-818abf54c3ad/application","descriptionHtml":"Quadrants is an open-source translation layer that turns ordinary Python into optimized machine code at runtime, across CPU (Arm64, x86) and GPU (AMDGPU, CUDA, Metal, Vulkan) backends. CUDA is the primary target, with x86 the strong runner-up, especially for real-time simulation. The others drive adoption and support fast prototyping for research.
The ambition is explicit: build the best “Python compiler” in the world for physics and performance-critical numerical computing. We are fully committed to it, and we intend to beat leading alternatives like Nvidia Warp, Numba, or Triton.
It’s a standalone project with its own roadmap, built in the open. Genesis, our physics simulator, is its first and top-priority user, the source of its hardest workloads and its benchmarks. But Quadrants is general-purpose infrastructure for anyone writing performance-critical numerical Python, not an internal tool that happens to be public.
The bet: high-performance code should feel natural to write. Engineers write plain Python in a NumPy or PyTorch style, and Quadrants makes it fast.
You own Quadrants end to end: the API, the IR and lowering, the optimization passes, the numerical primitives, and the roadmap.
Our ultimate objective: engineers write for readability and maintainability, never for performance. They write the version of an algorithm they are most comfortable with, and Quadrants rebuilds it into kernels that run as if an expert had hand-tuned them. Performance is the compiler’s job, not theirs. Getting there is a frontier problem, part research and part engineering, and we intend to define the state of the art on it, relentlessly pushing the limit further.
It calls for two things at once: the low-level GPU and compiler skill to make a single kernel optimal, and the ability to keep the whole system coherent as it grows.
The core tradeoff is settled: usability comes first. When usability and performance collide, usability wins, unless the cost is severe. We will limit what the system supports to go fast (static shapes and fixed-size allocation baked into kernels, say), but never the convenience of the API itself. Experts who want maximum performance and full control over the generated code get lower-level escape hatches, down to inline assembly, C, or PTX.
Your mandate is to build the best open-source Python compiler for physics simulation out there. No limit on what that takes. If it means replacing the entire engine, your call. Either way, you keep Quadrants healthy and unblocking the company in the meantime.
API & authoring experience. The surface that physics and numerical-computing engineers write against, and, as an open-source library, the product’s face to the world. Idiomatic, natural to Python users, never a separate language bolted onto Python. Clean decorators, legible constraints, sharp errors, real escape hatches.
IR, lowering & backends. A stable IR that lowers to multiple backends, with room to grow and the door kept open for differentiable simulation. Today the computation graph is expressed through Quadrants’ own constructs. Recovering it automatically from ordinary Python is hard, and a long-term goal rather than today’s reality.
Optimization. Turn naive code into strong execution plans: vectorization, register and shared-memory usage, barriers and atomics, cooperative threading, dead-buffer and redundant-traffic elimination, dropping needless synchronization. Move low-level optimization off users and into the platform.
Primitives. Cholesky, eigendecomposition, matmul, reductions, solver kernels. Each owned with its correctness, benchmarks, and regression tracking.
Watching Genesis and the compiler together. Hunt for performance wins on both sides, in Genesis and in Quadrants itself. Propose API changes that cut duplication, improve readability, and reduce platform-specific special-casing. Less to maintain, more that simply works.
The project as a product. Benchmark real workloads, track performance over time, catch regressions early. Stay close to users (Genesis first), find friction and missing abstractions, drive adoption, set direction.
You reason about whole systems and have taste for where complexity should live. You measure before you optimize, and you reach for memory traffic and launch overhead before FLOPs. You have opinions about API design. You write tests and benchmarks as you go, not afterward.
The ground that judgment stands on:
GPU. What optimal looks like and why: access patterns, registers, shared memory, barriers, atomics, threading. CPU intuition a plus.
Python & API taste. Idiomatic Python. A feel for what’s natural to Python users. APIs that free users rather than constrain them.
Compilers. The path from authoring to kernels (IR, lowering, optimization), weighed against maintainability.
And how you work: proactive and largely self-directed. You fetch context across the company instead of waiting for it. You’re comfortable talking to the engineers who use what you build, and at ease designing in the open for users you’ll never meet. Ready to grow into leading a small team (a couple of people, near term).
People adopt Quadrants from outside Genesis. Engineers write clearer code with less boilerplate. Kernels keep getting faster. Performance is tracked, not guessed. The architecture gets easier to extend, not harder.
How you work is up to you. What matters is that your ownership grows over time: blind spots shrink, the parts you can vouch for expand, technical debt goes down, the core stays sound.
That takes treating understanding as real work, not overhead. You read and challenge the codebase instead of just stacking features on it. You take nothing you did not write for granted. You watch the competition and what’s coming next, and you bring it in and ship it.
Use AI agents if they help. They are productivity tools, not code owners. You are.
","descriptionPlain":"WHAT QUADRANTS IS\n\nQuadrants is an open-source translation layer that turns ordinary Python into optimized machine code at runtime, across CPU (Arm64, x86) and GPU (AMDGPU, CUDA, Metal, Vulkan) backends. CUDA is the primary target, with x86 the strong runner-up, especially for real-time simulation. The others drive adoption and support fast prototyping for research.\n\nThe ambition is explicit: build the best “Python compiler” in the world for physics and performance-critical numerical computing. We are fully committed to it, and we intend to beat leading alternatives like Nvidia Warp, Numba, or Triton.\n\nIt’s a standalone project with its own roadmap, built in the open. Genesis, our physics simulator, is its first and top-priority user, the source of its hardest workloads and its benchmarks. But Quadrants is general-purpose infrastructure for anyone writing performance-critical numerical Python, not an internal tool that happens to be public.\n\nThe bet: high-performance code should feel natural to write. Engineers write plain Python in a NumPy or PyTorch style, and Quadrants makes it fast.\n\n\n\n\nTHE ROLE\n\nYou own Quadrants end to end: the API, the IR and lowering, the optimization passes, the numerical primitives, and the roadmap.\n\nOur ultimate objective: engineers write for readability and maintainability, never for performance. They write the version of an algorithm they are most comfortable with, and Quadrants rebuilds it into kernels that run as if an expert had hand-tuned them. Performance is the compiler’s job, not theirs. Getting there is a frontier problem, part research and part engineering, and we intend to define the state of the art on it, relentlessly pushing the limit further.\n\nIt calls for two things at once: the low-level GPU and compiler skill to make a single kernel optimal, and the ability to keep the whole system coherent as it grows.\n\nThe core tradeoff is settled: usability comes first. When usability and performance collide, usability wins, unless the cost is severe. We will limit what the system supports to go fast (static shapes and fixed-size allocation baked into kernels, say), but never the convenience of the API itself. Experts who want maximum performance and full control over the generated code get lower-level escape hatches, down to inline assembly, C, or PTX.\n\nYour mandate is to build the best open-source Python compiler for physics simulation out there. No limit on what that takes. If it means replacing the entire engine, your call. Either way, you keep Quadrants healthy and unblocking the company in the meantime.\n\n\n\n\nWHAT YOU’LL OWN\n\n - API & authoring experience. The surface that physics and numerical-computing engineers write against, and, as an open-source library, the product’s face to the world. Idiomatic, natural to Python users, never a separate language bolted onto Python. Clean decorators, legible constraints, sharp errors, real escape hatches.\n\n - IR, lowering & backends. A stable IR that lowers to multiple backends, with room to grow and the door kept open for differentiable simulation. Today the computation graph is expressed through Quadrants’ own constructs. Recovering it automatically from ordinary Python is hard, and a long-term goal rather than today’s reality.\n\n - Optimization. Turn naive code into strong execution plans: vectorization, register and shared-memory usage, barriers and atomics, cooperative threading, dead-buffer and redundant-traffic elimination, dropping needless synchronization. Move low-level optimization off users and into the platform.\n\n - Primitives. Cholesky, eigendecomposition, matmul, reductions, solver kernels. Each owned with its correctness, benchmarks, and regression tracking.\n\n - Watching Genesis and the compiler together. Hunt for performance wins on both sides, in Genesis and in Quadrants itself. Propose API changes that cut duplication, improve readability, and reduce platform-specific special-casing. Less to maintain, more that simply works.\n\n - The project as a product. Benchmark real workloads, track performance over time, catch regressions early. Stay close to users (Genesis first), find friction and missing abstractions, drive adoption, set direction.\n\n\n\n\nWHO YOU ARE\n\nYou reason about whole systems and have taste for where complexity should live. You measure before you optimize, and you reach for memory traffic and launch overhead before FLOPs. You have opinions about API design. You write tests and benchmarks as you go, not afterward.\n\nThe ground that judgment stands on:\n\n - GPU. What optimal looks like and why: access patterns, registers, shared memory, barriers, atomics, threading. CPU intuition a plus.\n\n - Python & API taste. Idiomatic Python. A feel for what’s natural to Python users. APIs that free users rather than constrain them.\n\n - Compilers. The path from authoring to kernels (IR, lowering, optimization), weighed against maintainability.\n\nAnd how you work: proactive and largely self-directed. You fetch context across the company instead of waiting for it. You’re comfortable talking to the engineers who use what you build, and at ease designing in the open for users you’ll never meet. Ready to grow into leading a small team (a couple of people, near term).\n\n\n\n\nWHAT GREAT LOOKS LIKE\n\nPeople adopt Quadrants from outside Genesis. Engineers write clearer code with less boilerplate. Kernels keep getting faster. Performance is tracked, not guessed. The architecture gets easier to extend, not harder.\n\n\n\n\nWHAT WE’RE UNCOMPROMISING ABOUT\n\nHow you work is up to you. What matters is that your ownership grows over time: blind spots shrink, the parts you can vouch for expand, technical debt goes down, the core stays sound.\n\nThat takes treating understanding as real work, not overhead. You read and challenge the codebase instead of just stacking features on it. You take nothing you did not write for granted. You watch the competition and what’s coming next, and you bring it in and ship it.\n\nUse AI agents if they help. They are productivity tools, not code owners. You are."},{"id":"490c5852-217d-47b9-ae59-677c483ee9c6","title":"Genesis-World: Core Simulation Engine Engineer","department":"Engineering & Research","team":"Simulation","employmentType":"FullTime","location":"Bay Area","secondaryLocations":[],"publishedAt":"2026-08-19T00:43:14.222+00:00","isListed":true,"isRemote":false,"workplaceType":"OnSite","address":{"postalAddress":{"addressRegion":"California","addressCountry":"United States","addressLocality":"San Carlos"}},"jobUrl":"https://jobs.ashbyhq.com/genesis/490c5852-217d-47b9-ae59-677c483ee9c6","applyUrl":"https://jobs.ashbyhq.com/genesis/490c5852-217d-47b9-ae59-677c483ee9c6/application","descriptionHtml":"Robots will learn in simulation before they hit the factory. Genesis-World is our bet on that future.
Genesis-World is an open-source, general-purpose simulation platform for physical AI from Genesis AI. One unified multi-physics engine: rigid bodies, FEM, MPM, particles, cloth, fluids. A robot arm can pour water onto sand, grasp a deformable object, or cut a soft body, all in the same simulation. Nyx, our in-house renderer, may be the most promising renderer for robotics out there: real-time photo-realistic rendering, advanced features like depth of field, and state-of-the-art techniques never seen before. Sensors of every kind: cameras, lidar, IMU, contact forces, temperature, plus arguably the most advanced tactile simulation available (paper). And the engine keeps growing: we are developing internally the most comprehensive and fastest Incremental Potential Contact (paper) solver for deformable body dynamics we know of, soon to be open-sourced. It powers real business applications, from full-fledged box packaging with labelling machine and all, to wire harnessing and lab automation, without any physics hack or compromise.
Everything is Python-first and runs anywhere. Kernels are written once, and Quadrants, our in-house JIT compiler, lowers them to CUDA, AMD ROCm, Apple Metal, Vulkan, x86, and ARM64. A single laptop or a datacenter. Massively batched GPU simulation for learning at scale, and complex non-batched scenes where CPU wins outright.
This is at the core of Genesis AI’s strategy. Evaluation is the bottleneck of scalable robotics: real hardware caps iteration at wall-clock time, but simulation turns it into a compute problem. Ours already runs two orders of magnitude faster than hardware (tens of thousands of episodes in half an hour instead of 200+ hours), while correlating with on-hardware rollouts at 89%. The north star: physical AI that improves at the speed of compute.
Simulation is still a hard sell in robotics. Outside a few success stories, like reinforcement learning for locomotion, most teams skip it, and friction is a big part of why: painful to use, painful to debug. We live that pain ourselves: debugging a failing experiment during policy evaluation takes 30 minutes, and there is no way to automatically generate a standalone reproduction script. That blocks adoption, even internally.
Your job is to kill that friction. You build the software engineering backbone of Genesis-World: everything that makes a large, fast-moving simulation codebase reliable, maintainable, frictionless. And it all ships in the open. Every hour of pain you remove, you remove for every roboticist who comes after you.
You join as core maintainer of Genesis-World, with shared stewardship of the whole platform from day one. Our physics team develops the algorithms, but just as importantly, you work to make everything around them excellent: infrastructure, tooling, APIs, architecture.
Make testing of a JIT-compiled engine scalable. Running our full suite is already taking hours because the ratio between compile time and runtime can be as bad as x10. Getting rid of this bottleneck is a serious challenge that will require a joint effort from the Genesis-World and the compiler teams.
Make any bug reproducible. Dump and reload simulations bit-exactly, across machines, across backends, even when a single timestep of a batched simulation can already max out VRAM. Turn any failing run into a standalone reproduction script, automatically.
Make autodiff first-class. Genesis-World is differentiable today, but partially: memory-hungry, slower than it could be, and hard to maintain due to manual operation recording. The project spans the engine and the compiler: you drive the requirements on Quadrants and iterate in a tight loop with the compiler team until the integration is solid.
Edit scenes at frozen time. See the scene, move entities and robots around, iterate on layout, all without building the scene or compiling the engine. Authoring today means a full build cycle. Make it instant.
Make the simulator extensible. First-class plugins for physics solvers and interactive viewer backends, including Nyx. Sensors already work this way. Generalize it, and Genesis-World becomes a platform others build on.
Give robots real actuation. Motors and MIMO transmissions, cleanly abstracted at the interface between users and the physics engine.
Make time a local variable. Today the clock and the timestep are global to the engine, which forbids whole classes of speedups: environments cannot step independently, the timestep cannot adapt to the state, and individual solvers cannot advance at their own rate. Free the clock down to the sub-scene level and let the simulation fly.
Make scenes come alive. Events triggered by physics: press the button and the microwave door pops open, squeeze the trigger and the drill spins.
Keep the core sharp. Refactoring legacy subsystems, stronger typing, consistent coding style, less Python overhead on hot paths, full scene serialization, telemetry and replay.
Beyond the code: you review issues and pull requests with the core team, answer questions from users and contributors, and can join the push on adoption.
You care about the engineer using what you build as much as the thing itself. A slow test, a cryptic error, a manual step: you see it, you go after the root cause, and you remove it for everyone. You have opinions about API design.
You are comfortable with physics and robotics: the right design choices take a working picture of what a simulator like Genesis-World does, how it works, and how it is used, a picture you can partly build on the job.
Strong Python engineering on large codebases: architecture, refactoring, typing, packaging, performance profiling (py-spy, Nsight, Xcode Instruments, RenderDoc).
Developer tooling, test infrastructure, or CI at scale.
Enough fluency with JIT compilation, code generation, or GPU computing stacks to reason about compilation caching, kernel dispatch, and startup cost. You will work closely with Quadrants.
At ease debugging across the full platform matrix: Windows, Linux, and macOS, on x86 and arm64, over CUDA, AMD, and Apple Metal (and possibly Intel XPU via Vulkan soon)
Open-source reflexes: triaging issues, reviewing external pull requests, communicating with a community.
Bonus points: experience with other robotics simulators (Drake, MuJoCo, Newton, Isaac, RaiSim, Jiminy, Brax). PyTorch, our user-facing interface. Other JIT compilers such as Triton, JAX, or Numba. 3D tooling and interactive editors (Maya, Blender, Houdini). Serialization formats (USD, glTF, MJCF/URDF). C++, CUDA, or Metal. Developer relations or community work.
","descriptionPlain":"WHAT WE’RE BUILDING\n\nRobots will learn in simulation before they hit the factory. Genesis-World is our bet on that future.\n\nGenesis-World https://github.com/Genesis-Embodied-AI/genesis-world is an open-source, general-purpose simulation platform for physical AI from Genesis AI https://www.genesis.ai/. One unified multi-physics engine: rigid bodies, FEM, MPM, particles, cloth, fluids. A robot arm can pour water onto sand, grasp a deformable object, or cut a soft body, all in the same simulation. Nyx, our in-house renderer, may be the most promising renderer for robotics out there: real-time photo-realistic rendering, advanced features like depth of field, and state-of-the-art techniques never seen before. Sensors of every kind: cameras, lidar, IMU, contact forces, temperature, plus arguably the most advanced tactile simulation available (paper https://openreview.net/forum?id=md8q2kfZHP). And the engine keeps growing: we are developing internally the most comprehensive and fastest Incremental Potential Contact (paper https://ipc-sim.github.io) solver for deformable body dynamics we know of, soon to be open-sourced. It powers real business applications, from full-fledged box packaging with labelling machine and all, to wire harnessing and lab automation, without any physics hack or compromise.\n\nEverything is Python-first and runs anywhere. Kernels are written once, and Quadrants, our in-house JIT compiler, lowers them to CUDA, AMD ROCm, Apple Metal, Vulkan, x86, and ARM64. A single laptop or a datacenter. Massively batched GPU simulation for learning at scale, and complex non-batched scenes where CPU wins outright.\n\nThis is at the core of Genesis AI’s strategy https://www.genesis.ai/blog/the-role-of-simulation-in-scalable-robotics-genesis-world-10-and-the-path-forward. Evaluation is the bottleneck of scalable robotics: real hardware caps iteration at wall-clock time, but simulation turns it into a compute problem. Ours already runs two orders of magnitude faster than hardware (tens of thousands of episodes in half an hour instead of 200+ hours), while correlating with on-hardware rollouts at 89%. The north star: physical AI that improves at the speed of compute.\n\n\n\n\nTHE ROLE\n\nSimulation is still a hard sell in robotics. Outside a few success stories, like reinforcement learning for locomotion, most teams skip it, and friction is a big part of why: painful to use, painful to debug. We live that pain ourselves: debugging a failing experiment during policy evaluation takes 30 minutes, and there is no way to automatically generate a standalone reproduction script. That blocks adoption, even internally.\n\nYour job is to kill that friction. You build the software engineering backbone of Genesis-World: everything that makes a large, fast-moving simulation codebase reliable, maintainable, frictionless. And it all ships in the open. Every hour of pain you remove, you remove for every roboticist who comes after you.\n\nYou join as core maintainer of Genesis-World, with shared stewardship of the whole platform from day one. Our physics team develops the algorithms, but just as importantly, you work to make everything around them excellent: infrastructure, tooling, APIs, architecture.\n\n\n\n\nTHE PROBLEMS WAITING FOR YOU\n\n - Make testing of a JIT-compiled engine scalable. Running our full suite is already taking hours because the ratio between compile time and runtime can be as bad as x10. Getting rid of this bottleneck is a serious challenge that will require a joint effort from the Genesis-World and the compiler teams.\n\n - Make any bug reproducible. Dump and reload simulations bit-exactly, across machines, across backends, even when a single timestep of a batched simulation can already max out VRAM. Turn any failing run into a standalone reproduction script, automatically.\n\n - Make autodiff first-class. Genesis-World is differentiable today, but partially: memory-hungry, slower than it could be, and hard to maintain due to manual operation recording. The project spans the engine and the compiler: you drive the requirements on Quadrants and iterate in a tight loop with the compiler team until the integration is solid.\n\n - Edit scenes at frozen time. See the scene, move entities and robots around, iterate on layout, all without building the scene or compiling the engine. Authoring today means a full build cycle. Make it instant.\n\n - Make the simulator extensible. First-class plugins for physics solvers and interactive viewer backends, including Nyx. Sensors already work this way. Generalize it, and Genesis-World becomes a platform others build on.\n\n - Give robots real actuation. Motors and MIMO transmissions, cleanly abstracted at the interface between users and the physics engine.\n\n - Make time a local variable. Today the clock and the timestep are global to the engine, which forbids whole classes of speedups: environments cannot step independently, the timestep cannot adapt to the state, and individual solvers cannot advance at their own rate. Free the clock down to the sub-scene level and let the simulation fly.\n\n - Make scenes come alive. Events triggered by physics: press the button and the microwave door pops open, squeeze the trigger and the drill spins.\n\n - Keep the core sharp. Refactoring legacy subsystems, stronger typing, consistent coding style, less Python overhead on hot paths, full scene serialization, telemetry and replay.\n\nBeyond the code: you review issues and pull requests with the core team, answer questions from users and contributors, and can join the push on adoption.\n\n\n\n\nWHO YOU ARE\n\nYou care about the engineer using what you build as much as the thing itself. A slow test, a cryptic error, a manual step: you see it, you go after the root cause, and you remove it for everyone. You have opinions about API design.\n\nYou are comfortable with physics and robotics: the right design choices take a working picture of what a simulator like Genesis-World does, how it works, and how it is used, a picture you can partly build on the job.\n\n - Strong Python engineering on large codebases: architecture, refactoring, typing, packaging, performance profiling (py-spy, Nsight, Xcode Instruments, RenderDoc).\n\n - Developer tooling, test infrastructure, or CI at scale.\n\n - Enough fluency with JIT compilation, code generation, or GPU computing stacks to reason about compilation caching, kernel dispatch, and startup cost. You will work closely with Quadrants.\n\n - At ease debugging across the full platform matrix: Windows, Linux, and macOS, on x86 and arm64, over CUDA, AMD, and Apple Metal (and possibly Intel XPU via Vulkan soon)\n\n - Open-source reflexes: triaging issues, reviewing external pull requests, communicating with a community.\n\nBonus points: experience with other robotics simulators (Drake, MuJoCo, Newton, Isaac, RaiSim, Jiminy, Brax). PyTorch, our user-facing interface. Other JIT compilers such as Triton, JAX, or Numba. 3D tooling and interactive editors (Maya, Blender, Houdini). Serialization formats (USD, glTF, MJCF/URDF). C++, CUDA, or Metal. Developer relations or community work."},{"id":"a99e725f-fa13-4f41-9ffc-3ad6305eb4e1","title":"Genesis-World: Core Physics Engineer","department":"Engineering & Research","team":"Simulation","employmentType":"FullTime","location":"Bay Area","secondaryLocations":[],"publishedAt":"2026-08-19T00:45:30.499+00:00","isListed":true,"isRemote":false,"workplaceType":"OnSite","address":{"postalAddress":{"addressRegion":"California","addressCountry":"United States","addressLocality":"San Carlos"}},"jobUrl":"https://jobs.ashbyhq.com/genesis/a99e725f-fa13-4f41-9ffc-3ad6305eb4e1","applyUrl":"https://jobs.ashbyhq.com/genesis/a99e725f-fa13-4f41-9ffc-3ad6305eb4e1/application","descriptionHtml":"Robots will learn in simulation before they hit the factory. Genesis-World is our bet on that future.
Genesis-World is an open-source, general-purpose simulation platform for physical AI from Genesis AI. One unified multi-physics engine: rigid bodies, FEM, MPM, particles, cloth, fluids. A robot arm can pour water onto sand, grasp a deformable object, or cut a soft body, all in the same simulation. Nyx, our in-house renderer, may be the most promising renderer for robotics out there: real-time photo-realistic rendering, advanced features like depth of field, and state-of-the-art techniques never seen before. Sensors of every kind: cameras, lidar, IMU, contact forces, temperature, plus arguably the most advanced tactile simulation available (paper). And the engine keeps growing: we are developing internally the most comprehensive and fastest Incremental Potential Contact (paper) solver for deformable body dynamics we know of, soon to be open-sourced. It powers real business applications, from full-fledged box packaging with labelling machine and all, to wire harnessing and lab automation, without any physics hack or compromise.
Everything is Python-first and runs anywhere. Kernels are written once, and Quadrants, our in-house JIT compiler, lowers them to CUDA, AMD ROCm, Apple Metal, Vulkan, x86, and ARM64. A single laptop or a datacenter. Massively batched GPU simulation for learning at scale, and complex non-batched scenes where CPU wins outright.
This is at the core of Genesis AI's strategy. Evaluation is the bottleneck of scalable robotics: real hardware caps iteration at wall-clock time, but simulation turns it into a compute problem. Ours already runs two orders of magnitude faster than hardware (tens of thousands of episodes in half an hour instead of 200+ hours), while correlating with on-hardware rollouts at 89%. The north star: physical AI that improves at the speed of compute.
You push the physics of Genesis-World forward. The mandate is clear: ship production-ready simulation capabilities that matter for the company's internal needs. Research applied end-to-end, from algorithm to merged, tested, documented code that real robot-learning pipelines depend on. Occasional groundbreaking research happens, notably through academic collaborations. But the core of the job is making the engine measurably better along five axes:
Speed. Algorithms that are not only faster but also smart enough to spend compute only where it matters across both time and space: larger stable timesteps, selective fidelity (adaptive across scales or simply hand-set), structure-aware solvers.
Completeness. No physics off limits: water, human animation, air flow, gravel, tendons, even body organs. Whatever the next use-case needs, the engine grows to cover it.
Fidelity. More realistic models: contact, friction, deformation, energy, actuation, materials…
Versatility. Extensible multi-physics without compromise on realism: all solvers in the scene coupled together at once, two-way and constraint-based. Write your own solver and it joins the scene like a native one, growing into an open solver ecosystem.
Scalability. From workstation to factory scale, and one day, city scale: thousands of interacting entities, batched across environments, without losing physical soundness.
Our ambition is to establish Genesis-World as the go-to simulator for physical AI, from companies and research labs to individuals.
Every fidelity for every physics. The same physics at every point of the speed-accuracy spectrum, from heavily batched training with XPBD or VBD to final validation with IPC. Same scene, same API, pick your tradeoff.
Invent physics level-of-detail (LOD). Rendering has had LOD for decades, physics is still waiting. Simulate at full fidelity what agents interact with and see, coarsely what they do not.
Heterogeneous environments. Every parallel world can hold a completely different model: different bodies, joints, and collision geometries.
Adaptive timesteps per island. Error-based control with Runge-Kutta Dopri5, and Time-of-Impact stepping during collision detection, as done in Jiminy.
Couple everything, exactly. Efficient and accurate two-way constraint-based coupling between heterogeneous grey-box solvers, using state-of-the-art methods like ADMM. Owning every solver in the stack is what makes it possible.
More scalable constraint solvers. Push rigid constraint solving beyond its current scalability ceiling (reference).
Unify contact resolution. Hydro-elastic compliance, unilateral constraints, and sequential impulses in the same framework, ideally under one generic formulation.
Closed kinematic loops without constraints. Handle loops intrinsically for numerical stability and speed, in the spirit of Kamino.
Day to day: you write your physics in plain Python and Quadrants makes it fast on every backend. And you validate it the hard way: analytical closed forms, other engines, real-world data.
You are a physicist and an engineer at once. You judge a method by whether it holds up in production at real scale, and you do not stop until it does. No blind spots: you relentlessly hunt down even the defect that looks insignificant, because it never is.
A strong background in physics-based simulation, preferably related to robotics: RBD, FEM, MPM, SPH, IPC, XPBD, VBD, ABD, plus constrained optimization and numerical integration of stiff systems.
A track record of shipping simulation code that others rely on, in an engine, in industry, or in a research codebase used beyond its authors.
Solid HPC programming (CPU and/or GPU), and an instinct for what makes a numerical method fast in practice, beyond complexity classes.
Rigor in validation: analytical closed forms, cross-engine consistency, real-world data.
Bonus points: publications in simulation, graphics, or robotics venues (SIGGRAPH, ICRA, IROS, CoRL, RSS). Contributions to an open-source physics engine.
","descriptionPlain":"WHAT WE'RE BUILDING\n\nRobots will learn in simulation before they hit the factory. Genesis-World is our bet on that future.\n\nGenesis-World https://github.com/Genesis-Embodied-AI/genesis-world is an open-source, general-purpose simulation platform for physical AI from Genesis AI https://www.genesis.ai/. One unified multi-physics engine: rigid bodies, FEM, MPM, particles, cloth, fluids. A robot arm can pour water onto sand, grasp a deformable object, or cut a soft body, all in the same simulation. Nyx, our in-house renderer, may be the most promising renderer for robotics out there: real-time photo-realistic rendering, advanced features like depth of field, and state-of-the-art techniques never seen before. Sensors of every kind: cameras, lidar, IMU, contact forces, temperature, plus arguably the most advanced tactile simulation available (paper https://openreview.net/forum?id=md8q2kfZHP). And the engine keeps growing: we are developing internally the most comprehensive and fastest Incremental Potential Contact (paper https://ipc-sim.github.io) solver for deformable body dynamics we know of, soon to be open-sourced. It powers real business applications, from full-fledged box packaging with labelling machine and all, to wire harnessing and lab automation, without any physics hack or compromise.\n\nEverything is Python-first and runs anywhere. Kernels are written once, and Quadrants, our in-house JIT compiler, lowers them to CUDA, AMD ROCm, Apple Metal, Vulkan, x86, and ARM64. A single laptop or a datacenter. Massively batched GPU simulation for learning at scale, and complex non-batched scenes where CPU wins outright.\n\nThis is at the core of Genesis AI's strategy https://www.genesis.ai/blog/the-role-of-simulation-in-scalable-robotics-genesis-world-10-and-the-path-forward. Evaluation is the bottleneck of scalable robotics: real hardware caps iteration at wall-clock time, but simulation turns it into a compute problem. Ours already runs two orders of magnitude faster than hardware (tens of thousands of episodes in half an hour instead of 200+ hours), while correlating with on-hardware rollouts at 89%. The north star: physical AI that improves at the speed of compute.\n\n\n\n\nTHE ROLE\n\nYou push the physics of Genesis-World forward. The mandate is clear: ship production-ready simulation capabilities that matter for the company's internal needs. Research applied end-to-end, from algorithm to merged, tested, documented code that real robot-learning pipelines depend on. Occasional groundbreaking research happens, notably through academic collaborations. But the core of the job is making the engine measurably better along five axes:\n\n - Speed. Algorithms that are not only faster but also smart enough to spend compute only where it matters across both time and space: larger stable timesteps, selective fidelity (adaptive across scales or simply hand-set), structure-aware solvers.\n\n - Completeness. No physics off limits: water, human animation, air flow, gravel, tendons, even body organs. Whatever the next use-case needs, the engine grows to cover it.\n\n - Fidelity. More realistic models: contact, friction, deformation, energy, actuation, materials…\n\n - Versatility. Extensible multi-physics without compromise on realism: all solvers in the scene coupled together at once, two-way and constraint-based. Write your own solver and it joins the scene like a native one, growing into an open solver ecosystem.\n\n - Scalability. From workstation to factory scale, and one day, city scale: thousands of interacting entities, batched across environments, without losing physical soundness.\n\nOur ambition is to establish Genesis-World as the go-to simulator for physical AI, from companies and research labs to individuals.\n\n\n\n\nTHE PROBLEMS WAITING FOR YOU\n\n - Every fidelity for every physics. The same physics at every point of the speed-accuracy spectrum, from heavily batched training with XPBD or VBD to final validation with IPC. Same scene, same API, pick your tradeoff.\n\n - Invent physics level-of-detail (LOD). Rendering has had LOD for decades, physics is still waiting. Simulate at full fidelity what agents interact with and see, coarsely what they do not.\n\n - Heterogeneous environments. Every parallel world can hold a completely different model: different bodies, joints, and collision geometries.\n\n - Adaptive timesteps per island. Error-based control with Runge-Kutta Dopri5, and Time-of-Impact stepping during collision detection, as done in Jiminy.\n\n - Couple everything, exactly. Efficient and accurate two-way constraint-based coupling between heterogeneous grey-box solvers, using state-of-the-art methods like ADMM. Owning every solver in the stack is what makes it possible.\n\n - More scalable constraint solvers. Push rigid constraint solving beyond its current scalability ceiling (reference https://www.roboticsproceedings.org/rss20/p108.pdf).\n\n - Unify contact resolution. Hydro-elastic compliance, unilateral constraints, and sequential impulses in the same framework, ideally under one generic formulation.\n\n - Closed kinematic loops without constraints. Handle loops intrinsically for numerical stability and speed, in the spirit of Kamino https://disneyresearch.github.io/kamino.\n\nDay to day: you write your physics in plain Python and Quadrants makes it fast on every backend. And you validate it the hard way: analytical closed forms, other engines, real-world data.\n\n\n\n\nWHO YOU ARE\n\nYou are a physicist and an engineer at once. You judge a method by whether it holds up in production at real scale, and you do not stop until it does. No blind spots: you relentlessly hunt down even the defect that looks insignificant, because it never is.\n\n - A strong background in physics-based simulation, preferably related to robotics: RBD, FEM, MPM, SPH, IPC, XPBD, VBD, ABD, plus constrained optimization and numerical integration of stiff systems.\n\n - A track record of shipping simulation code that others rely on, in an engine, in industry, or in a research codebase used beyond its authors.\n\n - Solid HPC programming (CPU and/or GPU), and an instinct for what makes a numerical method fast in practice, beyond complexity classes.\n\n - Rigor in validation: analytical closed forms, cross-engine consistency, real-world data.\n\nBonus points: publications in simulation, graphics, or robotics venues (SIGGRAPH, ICRA, IROS, CoRL, RSS). Contributions to an open-source physics engine."},{"id":"cfb8445f-851b-4aeb-ac73-01323a1d8c45","title":"Quadrants: Compiler Lead","department":"Engineering & Research","team":"Simulation","employmentType":"FullTime","location":"Bay Area","secondaryLocations":[],"publishedAt":"2026-08-19T00:46:49.930+00:00","isListed":true,"isRemote":false,"workplaceType":"OnSite","address":{"postalAddress":{"addressRegion":"California","addressCountry":"United States","addressLocality":"San Carlos"}},"jobUrl":"https://jobs.ashbyhq.com/genesis/cfb8445f-851b-4aeb-ac73-01323a1d8c45","applyUrl":"https://jobs.ashbyhq.com/genesis/cfb8445f-851b-4aeb-ac73-01323a1d8c45/application","descriptionHtml":"Quadrants is an open-source translation layer that turns ordinary Python into optimized machine code at runtime, across CPU (Arm64, x86) and GPU (AMDGPU, CUDA, Metal, Vulkan) backends. CUDA is the primary target, with x86 the strong runner-up, especially for real-time simulation. The others drive adoption and support fast prototyping for research.
The ambition is explicit: build the best “Python compiler” in the world for physics and performance-critical numerical computing. We are fully committed to it, and we intend to beat leading alternatives like Nvidia Warp, Numba, or Triton.
It’s a standalone project with its own roadmap, built in the open. Genesis, our physics simulator, is its first and top-priority user, the source of its hardest workloads and its benchmarks. But Quadrants is general-purpose infrastructure for anyone writing performance-critical numerical Python, not an internal tool that happens to be public.
The bet: high-performance code should feel natural to write. Engineers write plain Python in a NumPy or PyTorch style, and Quadrants makes it fast.
You own Quadrants end to end: the API, the IR and lowering, the optimization passes, the numerical primitives, and the roadmap.
Our ultimate objective: engineers write for readability and maintainability, never for performance. They write the version of an algorithm they are most comfortable with, and Quadrants rebuilds it into kernels that run as if an expert had hand-tuned them. Performance is the compiler’s job, not theirs. Getting there is a frontier problem, part research and part engineering, and we intend to define the state of the art on it, relentlessly pushing the limit further.
It calls for two things at once: the low-level GPU and compiler skill to make a single kernel optimal, and the ability to keep the whole system coherent as it grows.
The core tradeoff is settled: usability comes first. When usability and performance collide, usability wins, unless the cost is severe. We will limit what the system supports to go fast (static shapes and fixed-size allocation baked into kernels, say), but never the convenience of the API itself. Experts who want maximum performance and full control over the generated code get lower-level escape hatches, down to inline assembly, C, or PTX.
Your mandate is to build the best open-source Python compiler for physics simulation out there. No limit on what that takes. If it means replacing the entire engine, your call. Either way, you keep Quadrants healthy and unblocking the company in the meantime.
API & authoring experience. The surface that physics and numerical-computing engineers write against, and, as an open-source library, the product’s face to the world. Idiomatic, natural to Python users, never a separate language bolted onto Python. Clean decorators, legible constraints, sharp errors, real escape hatches.
IR, lowering & backends. A stable IR that lowers to multiple backends, with room to grow and the door kept open for differentiable simulation. Today the computation graph is expressed through Quadrants’ own constructs. Recovering it automatically from ordinary Python is hard, and a long-term goal rather than today’s reality.
Optimization. Turn naive code into strong execution plans: vectorization, register and shared-memory usage, barriers and atomics, cooperative threading, dead-buffer and redundant-traffic elimination, dropping needless synchronization. Move low-level optimization off users and into the platform.
Primitives. Cholesky, eigendecomposition, matmul, reductions, solver kernels. Each owned with its correctness, benchmarks, and regression tracking.
Watching Genesis and the compiler together. Hunt for performance wins on both sides, in Genesis and in Quadrants itself. Propose API changes that cut duplication, improve readability, and reduce platform-specific special-casing. Less to maintain, more that simply works.
The project as a product. Benchmark real workloads, track performance over time, catch regressions early. Stay close to users (Genesis first), find friction and missing abstractions, drive adoption, set direction.
You reason about whole systems and have taste for where complexity should live. You measure before you optimize, and you reach for memory traffic and launch overhead before FLOPs. You have opinions about API design. You write tests and benchmarks as you go, not afterward.
The ground that judgment stands on:
GPU. What optimal looks like and why: access patterns, registers, shared memory, barriers, atomics, threading. CPU intuition a plus.
Python & API taste. Idiomatic Python. A feel for what’s natural to Python users. APIs that free users rather than constrain them.
Compilers. The path from authoring to kernels (IR, lowering, optimization), weighed against maintainability.
And how you work: proactive and largely self-directed. You fetch context across the company instead of waiting for it. You’re comfortable talking to the engineers who use what you build, and at ease designing in the open for users you’ll never meet. Ready to grow into leading a small team (a couple of people, near term).
People adopt Quadrants from outside Genesis. Engineers write clearer code with less boilerplate. Kernels keep getting faster. Performance is tracked, not guessed. The architecture gets easier to extend, not harder.
How you work is up to you. What matters is that your ownership grows over time: blind spots shrink, the parts you can vouch for expand, technical debt goes down, the core stays sound.
That takes treating understanding as real work, not overhead. You read and challenge the codebase instead of just stacking features on it. You take nothing you did not write for granted. You watch the competition and what’s coming next, and you bring it in and ship it.
Use AI agents if they help. They are productivity tools, not code owners. You are.
","descriptionPlain":"WHAT QUADRANTS IS\n\nQuadrants is an open-source translation layer that turns ordinary Python into optimized machine code at runtime, across CPU (Arm64, x86) and GPU (AMDGPU, CUDA, Metal, Vulkan) backends. CUDA is the primary target, with x86 the strong runner-up, especially for real-time simulation. The others drive adoption and support fast prototyping for research.\n\nThe ambition is explicit: build the best “Python compiler” in the world for physics and performance-critical numerical computing. We are fully committed to it, and we intend to beat leading alternatives like Nvidia Warp, Numba, or Triton.\n\nIt’s a standalone project with its own roadmap, built in the open. Genesis, our physics simulator, is its first and top-priority user, the source of its hardest workloads and its benchmarks. But Quadrants is general-purpose infrastructure for anyone writing performance-critical numerical Python, not an internal tool that happens to be public.\n\nThe bet: high-performance code should feel natural to write. Engineers write plain Python in a NumPy or PyTorch style, and Quadrants makes it fast.\n\n\n\n\nTHE ROLE\n\nYou own Quadrants end to end: the API, the IR and lowering, the optimization passes, the numerical primitives, and the roadmap.\n\nOur ultimate objective: engineers write for readability and maintainability, never for performance. They write the version of an algorithm they are most comfortable with, and Quadrants rebuilds it into kernels that run as if an expert had hand-tuned them. Performance is the compiler’s job, not theirs. Getting there is a frontier problem, part research and part engineering, and we intend to define the state of the art on it, relentlessly pushing the limit further.\n\nIt calls for two things at once: the low-level GPU and compiler skill to make a single kernel optimal, and the ability to keep the whole system coherent as it grows.\n\nThe core tradeoff is settled: usability comes first. When usability and performance collide, usability wins, unless the cost is severe. We will limit what the system supports to go fast (static shapes and fixed-size allocation baked into kernels, say), but never the convenience of the API itself. Experts who want maximum performance and full control over the generated code get lower-level escape hatches, down to inline assembly, C, or PTX.\n\nYour mandate is to build the best open-source Python compiler for physics simulation out there. No limit on what that takes. If it means replacing the entire engine, your call. Either way, you keep Quadrants healthy and unblocking the company in the meantime.\n\n\n\n\nWHAT YOU’LL OWN\n\n - API & authoring experience. The surface that physics and numerical-computing engineers write against, and, as an open-source library, the product’s face to the world. Idiomatic, natural to Python users, never a separate language bolted onto Python. Clean decorators, legible constraints, sharp errors, real escape hatches.\n\n - IR, lowering & backends. A stable IR that lowers to multiple backends, with room to grow and the door kept open for differentiable simulation. Today the computation graph is expressed through Quadrants’ own constructs. Recovering it automatically from ordinary Python is hard, and a long-term goal rather than today’s reality.\n\n - Optimization. Turn naive code into strong execution plans: vectorization, register and shared-memory usage, barriers and atomics, cooperative threading, dead-buffer and redundant-traffic elimination, dropping needless synchronization. Move low-level optimization off users and into the platform.\n\n - Primitives. Cholesky, eigendecomposition, matmul, reductions, solver kernels. Each owned with its correctness, benchmarks, and regression tracking.\n\n - Watching Genesis and the compiler together. Hunt for performance wins on both sides, in Genesis and in Quadrants itself. Propose API changes that cut duplication, improve readability, and reduce platform-specific special-casing. Less to maintain, more that simply works.\n\n - The project as a product. Benchmark real workloads, track performance over time, catch regressions early. Stay close to users (Genesis first), find friction and missing abstractions, drive adoption, set direction.\n\n\n\n\nWHO YOU ARE\n\nYou reason about whole systems and have taste for where complexity should live. You measure before you optimize, and you reach for memory traffic and launch overhead before FLOPs. You have opinions about API design. You write tests and benchmarks as you go, not afterward.\n\nThe ground that judgment stands on:\n\n - GPU. What optimal looks like and why: access patterns, registers, shared memory, barriers, atomics, threading. CPU intuition a plus.\n\n - Python & API taste. Idiomatic Python. A feel for what’s natural to Python users. APIs that free users rather than constrain them.\n\n - Compilers. The path from authoring to kernels (IR, lowering, optimization), weighed against maintainability.\n\nAnd how you work: proactive and largely self-directed. You fetch context across the company instead of waiting for it. You’re comfortable talking to the engineers who use what you build, and at ease designing in the open for users you’ll never meet. Ready to grow into leading a small team (a couple of people, near term).\n\n\n\n\nWHAT GREAT LOOKS LIKE\n\nPeople adopt Quadrants from outside Genesis. Engineers write clearer code with less boilerplate. Kernels keep getting faster. Performance is tracked, not guessed. The architecture gets easier to extend, not harder.\n\n\n\n\nWHAT WE’RE UNCOMPROMISING ABOUT\n\nHow you work is up to you. What matters is that your ownership grows over time: blind spots shrink, the parts you can vouch for expand, technical debt goes down, the core stays sound.\n\nThat takes treating understanding as real work, not overhead. You read and challenge the codebase instead of just stacking features on it. You take nothing you did not write for granted. You watch the competition and what’s coming next, and you bring it in and ship it.\n\nUse AI agents if they help. They are productivity tools, not code owners. You are."},{"id":"53919a1c-cfb9-46f2-831f-f5c6bbedf0af","title":"Robot Teleoperation Specialist","department":"Engineering & Research","team":"Robotics","employmentType":"Contract","location":"London","secondaryLocations":[],"publishedAt":"2026-09-07T15:36:56.951+00:00","isListed":true,"isRemote":true,"workplaceType":"Hybrid","address":{"postalAddress":{"addressCountry":"United Kingdom","addressLocality":"London"}},"jobUrl":"https://jobs.ashbyhq.com/genesis/53919a1c-cfb9-46f2-831f-f5c6bbedf0af","applyUrl":"https://jobs.ashbyhq.com/genesis/53919a1c-cfb9-46f2-831f-f5c6bbedf0af/application","descriptionHtml":"The role is part-time and contractor-based (1–2 days a week), with the option to ramp up. Minimum duration is 3 months, renewable. A great fit for a student.
Role Overview
Genesis is a global physical AI lab and full-stack robotics company. We build generalist robots to unlock unlimited physical labor, so humans can focus on creativity, curiosity, and what we love.
As a Robot Teleoperation Specialist, your primary responsibility is to operate robots during data collection sessions using teleoperation equipment. You’ll execute predefined tasks by following standard operating procedures, monitor robot behavior during operation, and ensure sessions are run safely and consistently.
This is a highly hands-on, execution-focused role. Technical setup, hardware maintenance, and data infrastructure are handled by other teams — when you arrive, everything is ready for you to teleoperate.
This role is onsite in London, UK.
Responsibilities
Operate robots using teleoperation equipment according to predefined tasks and SOPs
Execute teleoperation sessions reliably and consistently to support data collection
Observe robot behavior during operation and flag unexpected or unsafe behavior
Follow all safety guidelines to maintain a safe environment for yourself and the robot
Provide basic session feedback or notes when required
Maintain focus and quality during repeated teleoperation tasks
Collaborate with operations staff and follow direction from session leads
Requirements
Strong attention to detail and ability to follow instructions precisely
Comfortable working on your feet and using teleoperation or VR-style equipment
High consistency and reliability in executing repetitive tasks
Ability to stay focused for extended teleoperation sessions
Clear written and verbal communication skills
Coachable, curious, and open to feedback
Adaptable to changing task requirements in an R&D environment
Bonus Qualifications
Prior experience with robot teleoperation, simulation, or VR systems
Strong hand-eye coordination and spatial awareness
Ability to learn new physical tasks quickly and improve execution over time
Quadrants is an open-source translation layer that turns ordinary Python into optimized machine code at runtime, across CPU (Arm64, x86) and GPU (AMDGPU, CUDA, Metal, Vulkan) backends. CUDA is the primary target, with x86 the strong runner-up, especially for real-time simulation. The others drive adoption and support fast prototyping for research.
The ambition is explicit: build the best “Python compiler” in the world for physics and performance-critical numerical computing. We are fully committed to it, and we intend to beat leading alternatives like Nvidia Warp, Numba, or Triton.
It’s a standalone project with its own roadmap, built in the open. Genesis, our physics simulator, is its first and top-priority user, the source of its hardest workloads and its benchmarks. But Quadrants is general-purpose infrastructure for anyone writing performance-critical numerical Python, not an internal tool that happens to be public.
The bet: high-performance code should feel natural to write. Engineers write plain Python in a NumPy or PyTorch style, and Quadrants makes it fast.
You own Quadrants end to end: the API, the IR and lowering, the optimization passes, the numerical primitives, and the roadmap.
Our ultimate objective: engineers write for readability and maintainability, never for performance. They write the version of an algorithm they are most comfortable with, and Quadrants rebuilds it into kernels that run as if an expert had hand-tuned them. Performance is the compiler’s job, not theirs. Getting there is a frontier problem, part research and part engineering, and we intend to define the state of the art on it, relentlessly pushing the limit further.
It calls for two things at once: the low-level GPU and compiler skill to make a single kernel optimal, and the ability to keep the whole system coherent as it grows.
The core tradeoff is settled: usability comes first. When usability and performance collide, usability wins, unless the cost is severe. We will limit what the system supports to go fast (static shapes and fixed-size allocation baked into kernels, say), but never the convenience of the API itself. Experts who want maximum performance and full control over the generated code get lower-level escape hatches, down to inline assembly, C, or PTX.
API & authoring experience. The surface that physics and numerical-computing engineers write against, and, as an open-source library, the product’s face to the world. Idiomatic, natural to Python users, never a separate language bolted onto Python. Clean decorators, legible constraints, sharp errors, real escape hatches.
IR, lowering & backends. A stable IR that lowers to multiple backends, with room to grow and the door kept open for differentiable simulation. Today the computation graph is expressed through Quadrants’ own constructs. Recovering it automatically from ordinary Python is hard, and a long-term goal rather than today’s reality.
Optimization. Turn naive code into strong execution plans: vectorization, register and shared-memory usage, barriers and atomics, cooperative threading, dead-buffer and redundant-traffic elimination, dropping needless synchronization. Move low-level optimization off users and into the platform.
Primitives. Cholesky, eigendecomposition, matmul, reductions, solver kernels. Each owned with its correctness, benchmarks, and regression tracking.
Watching Genesis and the compiler together. Hunt for performance wins on both sides, in Genesis and in Quadrants itself. Propose API changes that cut duplication, improve readability, and reduce platform-specific special-casing. Less to maintain, more that simply works.
The project as a product. Benchmark real workloads, track performance over time, catch regressions early. Stay close to users (Genesis first), find friction and missing abstractions, drive adoption, set direction.
You reason about whole systems and have taste for where complexity should live. You measure before you optimize, and you reach for memory traffic and launch overhead before FLOPs. You have opinions about API design. You write tests and benchmarks as you go, not afterward.
The ground that judgment stands on:
GPU. What optimal looks like and why: access patterns, registers, shared memory, barriers, atomics, threading. CPU intuition a plus.
Python & API taste. Idiomatic Python. A feel for what’s natural to Python users. APIs that free users rather than constrain them.
Compilers. The path from authoring to kernels (IR, lowering, optimization), weighed against maintainability.
And how you work: proactive and largely self-directed. You fetch context across the company instead of waiting for it. You’re comfortable talking to the engineers who use what you build, and at ease designing in the open for users you’ll never meet. Ready to grow into leading a small team (a couple of people, near term).
People adopt Quadrants from outside Genesis. Engineers write clearer code with less boilerplate. Kernels keep getting faster. Performance is tracked, not guessed. The architecture gets easier to extend, not harder.
How you work is up to you. What matters is that your ownership grows over time: blind spots shrink, the parts you can vouch for expand, technical debt goes down, the core stays sound.
That takes treating understanding as real work, not overhead. You read and challenge the codebase instead of just stacking features on it. You take nothing you did not write for granted. You watch the competition and what’s coming next, and you bring it in and ship it.
Use AI agents if they help. They are productivity tools, not code owners. You are.
","descriptionPlain":"WHAT QUADRANTS IS\n\nQuadrants is an open-source translation layer that turns ordinary Python into optimized machine code at runtime, across CPU (Arm64, x86) and GPU (AMDGPU, CUDA, Metal, Vulkan) backends. CUDA is the primary target, with x86 the strong runner-up, especially for real-time simulation. The others drive adoption and support fast prototyping for research.\n\nThe ambition is explicit: build the best “Python compiler” in the world for physics and performance-critical numerical computing. We are fully committed to it, and we intend to beat leading alternatives like Nvidia Warp, Numba, or Triton.\n\nIt’s a standalone project with its own roadmap, built in the open. Genesis, our physics simulator, is its first and top-priority user, the source of its hardest workloads and its benchmarks. But Quadrants is general-purpose infrastructure for anyone writing performance-critical numerical Python, not an internal tool that happens to be public.\n\nThe bet: high-performance code should feel natural to write. Engineers write plain Python in a NumPy or PyTorch style, and Quadrants makes it fast.\n\n \n\n\nTHE ROLE\n\nYou own Quadrants end to end: the API, the IR and lowering, the optimization passes, the numerical primitives, and the roadmap.\n\nOur ultimate objective: engineers write for readability and maintainability, never for performance. They write the version of an algorithm they are most comfortable with, and Quadrants rebuilds it into kernels that run as if an expert had hand-tuned them. Performance is the compiler’s job, not theirs. Getting there is a frontier problem, part research and part engineering, and we intend to define the state of the art on it, relentlessly pushing the limit further.\n\nIt calls for two things at once: the low-level GPU and compiler skill to make a single kernel optimal, and the ability to keep the whole system coherent as it grows.\n\nThe core tradeoff is settled: usability comes first. When usability and performance collide, usability wins, unless the cost is severe. We will limit what the system supports to go fast (static shapes and fixed-size allocation baked into kernels, say), but never the convenience of the API itself. Experts who want maximum performance and full control over the generated code get lower-level escape hatches, down to inline assembly, C, or PTX.\n\n \n\n\nWHAT YOU’LL OWN\n\n - API & authoring experience. The surface that physics and numerical-computing engineers write against, and, as an open-source library, the product’s face to the world. Idiomatic, natural to Python users, never a separate language bolted onto Python. Clean decorators, legible constraints, sharp errors, real escape hatches.\n\n - IR, lowering & backends. A stable IR that lowers to multiple backends, with room to grow and the door kept open for differentiable simulation. Today the computation graph is expressed through Quadrants’ own constructs. Recovering it automatically from ordinary Python is hard, and a long-term goal rather than today’s reality.\n\n - Optimization. Turn naive code into strong execution plans: vectorization, register and shared-memory usage, barriers and atomics, cooperative threading, dead-buffer and redundant-traffic elimination, dropping needless synchronization. Move low-level optimization off users and into the platform.\n\n - Primitives. Cholesky, eigendecomposition, matmul, reductions, solver kernels. Each owned with its correctness, benchmarks, and regression tracking.\n\n - Watching Genesis and the compiler together. Hunt for performance wins on both sides, in Genesis and in Quadrants itself. Propose API changes that cut duplication, improve readability, and reduce platform-specific special-casing. Less to maintain, more that simply works.\n\n - The project as a product. Benchmark real workloads, track performance over time, catch regressions early. Stay close to users (Genesis first), find friction and missing abstractions, drive adoption, set direction.\n\n\nWHO YOU ARE\n\nYou reason about whole systems and have taste for where complexity should live. You measure before you optimize, and you reach for memory traffic and launch overhead before FLOPs. You have opinions about API design. You write tests and benchmarks as you go, not afterward.\n\nThe ground that judgment stands on:\n\n - GPU. What optimal looks like and why: access patterns, registers, shared memory, barriers, atomics, threading. CPU intuition a plus.\n\n - Python & API taste. Idiomatic Python. A feel for what’s natural to Python users. APIs that free users rather than constrain them.\n\n - Compilers. The path from authoring to kernels (IR, lowering, optimization), weighed against maintainability.\n\nAnd how you work: proactive and largely self-directed. You fetch context across the company instead of waiting for it. You’re comfortable talking to the engineers who use what you build, and at ease designing in the open for users you’ll never meet. Ready to grow into leading a small team (a couple of people, near term).\n\n \n\n\nWHAT GREAT LOOKS LIKE\n\nPeople adopt Quadrants from outside Genesis. Engineers write clearer code with less boilerplate. Kernels keep getting faster. Performance is tracked, not guessed. The architecture gets easier to extend, not harder.\n\n \n\n\nWHAT WE’RE UNCOMPROMISING ABOUT\n\nHow you work is up to you. What matters is that your ownership grows over time: blind spots shrink, the parts you can vouch for expand, technical debt goes down, the core stays sound.\n\nThat takes treating understanding as real work, not overhead. You read and challenge the codebase instead of just stacking features on it. You take nothing you did not write for granted. You watch the competition and what’s coming next, and you bring it in and ship it.\n\nUse AI agents if they help. They are productivity tools, not code owners. You are."},{"id":"d78b2039-2439-4cfb-889a-7c5fbd53976b","title":"Data Agent","department":"Engineering & Research","team":"Data","employmentType":"FullTime","location":"London","secondaryLocations":[],"publishedAt":"2026-09-07T14:59:50.992+00:00","isListed":true,"isRemote":true,"workplaceType":"Hybrid","address":{"postalAddress":{"addressCountry":"United Kingdom","addressLocality":"London"}},"jobUrl":"https://jobs.ashbyhq.com/genesis/d78b2039-2439-4cfb-889a-7c5fbd53976b","applyUrl":"https://jobs.ashbyhq.com/genesis/d78b2039-2439-4cfb-889a-7c5fbd53976b/application","descriptionHtml":"What You’ll Do
Develop LM/VLM-powered agents that autonomously interface with code, GUI, web tools, and design software to drive generative simulation
Build agentic systems that synthesize high-quality, diverse physical data by orchestrating simulation, generative models, and interactive tools
Build data pipelines to curate, generate, and preprocess datasets used for training and evaluating language- and vision-language-based agents
Collaborate with simulation and infrastructure teams to co-develop APIs, GUIs, and services that enable seamless interaction with physics engines and simulation environments
Drive innovation by staying at the forefront of LLM agents, autonomous tool use, and generative simulation — and translating cutting-edge research into robust, production-ready systems
Contribute to the creation of a generative simulation stack that empowers downstream applications in robotics, embodied AI, and physical reasoning
What You’ll Bring
A bold and imaginative vision for a next-generation paradigm of physical data synthesis— combining simulation, generative models, and autonomous agents
Deep curiosity and strong technical ownership, with a track record of driving complex, open-ended projects from concept to implementation
Experience with (multimodal) large language models, generative AI tools, agentic software design, GUI automation, program synthesis
Passion for inventing creative solutions at the edge of AI, simulation, and physical reasoning, with an eagerness to build systems that generate high-fidelity, useful data for embodied intelligence
Bonus: Familiarity with interactive design tools (e.g., CAD software, game engines) or simulation environments (e.g., physics engines, robotics simulators)
Robots will learn in simulation before they hit the factory. Genesis-World is our bet on that future.
Genesis-World is an open-source, general-purpose simulation platform for physical AI from Genesis AI. One unified multi-physics engine: rigid bodies, FEM, MPM, particles, cloth, fluids. A robot arm can pour water onto sand, grasp a deformable object, or cut a soft body, all in the same simulation. Nyx, our in-house renderer, may be the most promising renderer for robotics out there: real-time photo-realistic rendering, advanced features like depth of field, and state-of-the-art techniques never seen before. Sensors of every kind: cameras, lidar, IMU, contact forces, temperature, plus arguably the most advanced tactile simulation available (paper). And the engine keeps growing: we are developing internally the most comprehensive and fastest Incremental Potential Contact (paper) solver for deformable body dynamics we know of, soon to be open-sourced. It powers real business applications, from full-fledged box packaging with labelling machine and all, to wire harnessing and lab automation, without any physics hack or compromise.
Everything is Python-first and runs anywhere. Kernels are written once, and Quadrants, our in-house JIT compiler, lowers them to CUDA, AMD ROCm, Apple Metal, Vulkan, x86, and ARM64. A single laptop or a datacenter. Massively batched GPU simulation for learning at scale, and complex non-batched scenes where CPU wins outright.
This is at the core of Genesis AI’s strategy. Evaluation is the bottleneck of scalable robotics: real hardware caps iteration at wall-clock time, but simulation turns it into a compute problem. Ours already runs two orders of magnitude faster than hardware (tens of thousands of episodes in half an hour instead of 200+ hours), while correlating with on-hardware rollouts at 89%. The north star: physical AI that improves at the speed of compute.
Simulation is still a hard sell in robotics. Outside a few success stories, like reinforcement learning for locomotion, most teams skip it, and friction is a big part of why: painful to use, painful to debug. We live that pain ourselves: debugging a failing experiment during policy evaluation takes 30 minutes, and there is no way to automatically generate a standalone reproduction script. That blocks adoption, even internally.
Your job is to kill that friction. You build the software engineering backbone of Genesis-World: everything that makes a large, fast-moving simulation codebase reliable, maintainable, frictionless. And it all ships in the open. Every hour of pain you remove, you remove for every roboticist who comes after you.
You join as core maintainer of Genesis-World, with shared stewardship of the whole platform from day one. Our physics team develops the algorithms, but just as importantly, you work to make everything around them excellent: infrastructure, tooling, APIs, architecture.
Make testing of a JIT-compiled engine scalable. Running our full suite is already taking hours because the ratio between compile time and runtime can be as bad as x10. Getting rid of this bottleneck is a serious challenge that will require a joint effort from the Genesis-World and the compiler teams.
Make any bug reproducible. Dump and reload simulations bit-exactly, across machines, across backends, even when a single timestep of a batched simulation can already max out VRAM. Turn any failing run into a standalone reproduction script, automatically.
Make autodiff first-class. Genesis-World is differentiable today, but partially: memory-hungry, slower than it could be, and hard to maintain due to manual operation recording. The project spans the engine and the compiler: you drive the requirements on Quadrants and iterate in a tight loop with the compiler team until the integration is solid.
Edit scenes at frozen time. See the scene, move entities and robots around, iterate on layout, all without building the scene or compiling the engine. Authoring today means a full build cycle. Make it instant.
Make the simulator extensible. First-class plugins for physics solvers and interactive viewer backends, including Nyx. Sensors already work this way. Generalize it, and Genesis-World becomes a platform others build on.
Give robots real actuation. Motors and MIMO transmissions, cleanly abstracted at the interface between users and the physics engine.
Make time a local variable. Today the clock and the timestep are global to the engine, which forbids whole classes of speedups: environments cannot step independently, the timestep cannot adapt to the state, and individual solvers cannot advance at their own rate. Free the clock down to the sub-scene level and let the simulation fly.
Make scenes come alive. Events triggered by physics: press the button and the microwave door pops open, squeeze the trigger and the drill spins.
Keep the core sharp. Refactoring legacy subsystems, stronger typing, consistent coding style, less Python overhead on hot paths, full scene serialization, telemetry and replay.
Beyond the code: you review issues and pull requests with the core team, answer questions from users and contributors, and can join the push on adoption.
You care about the engineer using what you build as much as the thing itself. A slow test, a cryptic error, a manual step: you see it, you go after the root cause, and you remove it for everyone. You have opinions about API design.
You are comfortable with physics and robotics: the right design choices take a working picture of what a simulator like Genesis-World does, how it works, and how it is used, a picture you can partly build on the job.
Strong Python engineering on large codebases: architecture, refactoring, typing, packaging, performance profiling (py-spy, Nsight, Xcode Instruments, RenderDoc).
Developer tooling, test infrastructure, or CI at scale.
Enough fluency with JIT compilation, code generation, or GPU computing stacks to reason about compilation caching, kernel dispatch, and startup cost. You will work closely with Quadrants.
At ease debugging across the full platform matrix: Windows, Linux, and macOS, on x86 and arm64, over CUDA, AMD, and Apple Metal (and possibly Intel XPU via Vulkan soon)
Open-source reflexes: triaging issues, reviewing external pull requests, communicating with a community.
Bonus points: experience with other robotics simulators (Drake, MuJoCo, Newton, Isaac, RaiSim, Jiminy, Brax). PyTorch, our user-facing interface. Other JIT compilers such as Triton, JAX, or Numba. 3D tooling and interactive editors (Maya, Blender, Houdini). Serialization formats (USD, glTF, MJCF/URDF). C++, CUDA, or Metal. Developer relations or community work.
","descriptionPlain":"WHAT WE’RE BUILDING\n\nRobots will learn in simulation before they hit the factory. Genesis-World is our bet on that future.\n\nGenesis-World https://github.com/Genesis-Embodied-AI/genesis-world is an open-source, general-purpose simulation platform for physical AI from Genesis AI https://www.genesis.ai/. One unified multi-physics engine: rigid bodies, FEM, MPM, particles, cloth, fluids. A robot arm can pour water onto sand, grasp a deformable object, or cut a soft body, all in the same simulation. Nyx, our in-house renderer, may be the most promising renderer for robotics out there: real-time photo-realistic rendering, advanced features like depth of field, and state-of-the-art techniques never seen before. Sensors of every kind: cameras, lidar, IMU, contact forces, temperature, plus arguably the most advanced tactile simulation available (paper https://openreview.net/forum?id=md8q2kfZHP). And the engine keeps growing: we are developing internally the most comprehensive and fastest Incremental Potential Contact (paper https://ipc-sim.github.io) solver for deformable body dynamics we know of, soon to be open-sourced. It powers real business applications, from full-fledged box packaging with labelling machine and all, to wire harnessing and lab automation, without any physics hack or compromise.\n\nEverything is Python-first and runs anywhere. Kernels are written once, and Quadrants, our in-house JIT compiler, lowers them to CUDA, AMD ROCm, Apple Metal, Vulkan, x86, and ARM64. A single laptop or a datacenter. Massively batched GPU simulation for learning at scale, and complex non-batched scenes where CPU wins outright.\n\nThis is at the core of Genesis AI’s strategy https://www.genesis.ai/blog/the-role-of-simulation-in-scalable-robotics-genesis-world-10-and-the-path-forward. Evaluation is the bottleneck of scalable robotics: real hardware caps iteration at wall-clock time, but simulation turns it into a compute problem. Ours already runs two orders of magnitude faster than hardware (tens of thousands of episodes in half an hour instead of 200+ hours), while correlating with on-hardware rollouts at 89%. The north star: physical AI that improves at the speed of compute.\n\n\n\n\nTHE ROLE\n\nSimulation is still a hard sell in robotics. Outside a few success stories, like reinforcement learning for locomotion, most teams skip it, and friction is a big part of why: painful to use, painful to debug. We live that pain ourselves: debugging a failing experiment during policy evaluation takes 30 minutes, and there is no way to automatically generate a standalone reproduction script. That blocks adoption, even internally.\n\nYour job is to kill that friction. You build the software engineering backbone of Genesis-World: everything that makes a large, fast-moving simulation codebase reliable, maintainable, frictionless. And it all ships in the open. Every hour of pain you remove, you remove for every roboticist who comes after you.\n\nYou join as core maintainer of Genesis-World, with shared stewardship of the whole platform from day one. Our physics team develops the algorithms, but just as importantly, you work to make everything around them excellent: infrastructure, tooling, APIs, architecture.\n\n\n\n\nTHE PROBLEMS WAITING FOR YOU\n\n - Make testing of a JIT-compiled engine scalable. Running our full suite is already taking hours because the ratio between compile time and runtime can be as bad as x10. Getting rid of this bottleneck is a serious challenge that will require a joint effort from the Genesis-World and the compiler teams.\n\n - Make any bug reproducible. Dump and reload simulations bit-exactly, across machines, across backends, even when a single timestep of a batched simulation can already max out VRAM. Turn any failing run into a standalone reproduction script, automatically.\n\n - Make autodiff first-class. Genesis-World is differentiable today, but partially: memory-hungry, slower than it could be, and hard to maintain due to manual operation recording. The project spans the engine and the compiler: you drive the requirements on Quadrants and iterate in a tight loop with the compiler team until the integration is solid.\n\n - Edit scenes at frozen time. See the scene, move entities and robots around, iterate on layout, all without building the scene or compiling the engine. Authoring today means a full build cycle. Make it instant.\n\n - Make the simulator extensible. First-class plugins for physics solvers and interactive viewer backends, including Nyx. Sensors already work this way. Generalize it, and Genesis-World becomes a platform others build on.\n\n - Give robots real actuation. Motors and MIMO transmissions, cleanly abstracted at the interface between users and the physics engine.\n\n - Make time a local variable. Today the clock and the timestep are global to the engine, which forbids whole classes of speedups: environments cannot step independently, the timestep cannot adapt to the state, and individual solvers cannot advance at their own rate. Free the clock down to the sub-scene level and let the simulation fly.\n\n - Make scenes come alive. Events triggered by physics: press the button and the microwave door pops open, squeeze the trigger and the drill spins.\n\n - Keep the core sharp. Refactoring legacy subsystems, stronger typing, consistent coding style, less Python overhead on hot paths, full scene serialization, telemetry and replay.\n\nBeyond the code: you review issues and pull requests with the core team, answer questions from users and contributors, and can join the push on adoption.\n\n\n\n\nWHO YOU ARE\n\nYou care about the engineer using what you build as much as the thing itself. A slow test, a cryptic error, a manual step: you see it, you go after the root cause, and you remove it for everyone. You have opinions about API design.\n\nYou are comfortable with physics and robotics: the right design choices take a working picture of what a simulator like Genesis-World does, how it works, and how it is used, a picture you can partly build on the job.\n\n - Strong Python engineering on large codebases: architecture, refactoring, typing, packaging, performance profiling (py-spy, Nsight, Xcode Instruments, RenderDoc).\n\n - Developer tooling, test infrastructure, or CI at scale.\n\n - Enough fluency with JIT compilation, code generation, or GPU computing stacks to reason about compilation caching, kernel dispatch, and startup cost. You will work closely with Quadrants.\n\n - At ease debugging across the full platform matrix: Windows, Linux, and macOS, on x86 and arm64, over CUDA, AMD, and Apple Metal (and possibly Intel XPU via Vulkan soon)\n\n - Open-source reflexes: triaging issues, reviewing external pull requests, communicating with a community.\n\nBonus points: experience with other robotics simulators (Drake, MuJoCo, Newton, Isaac, RaiSim, Jiminy, Brax). PyTorch, our user-facing interface. Other JIT compilers such as Triton, JAX, or Numba. 3D tooling and interactive editors (Maya, Blender, Houdini). Serialization formats (USD, glTF, MJCF/URDF). C++, CUDA, or Metal. Developer relations or community work."},{"id":"2196f682-bd01-4f83-992d-367bbb3c8e8b","title":"Training / AI Infrastructure","department":"Engineering & Research","team":"Training","employmentType":"FullTime","location":"London","secondaryLocations":[],"publishedAt":"2026-09-07T15:57:24.750+00:00","isListed":true,"isRemote":true,"workplaceType":"Hybrid","address":{"postalAddress":{"addressCountry":"United Kingdom","addressLocality":"London"}},"jobUrl":"https://jobs.ashbyhq.com/genesis/2196f682-bd01-4f83-992d-367bbb3c8e8b","applyUrl":"https://jobs.ashbyhq.com/genesis/2196f682-bd01-4f83-992d-367bbb3c8e8b/application","descriptionHtml":"Drive down wall-clock time to convergence by profiling and eliminating bottlenecks across the foundation model training stack stack, from data pipelines to GPU kernels
Design, build, and optimize distributed training systems (PyTorch) for multi-node GPU clusters, ensuring scalability, robustness, and high utilization
Implement efficient low-level code (CUDA, cuDNN, Triton, custom kernels) and integrate it seamlessly into high-level training frameworks
Optimize workloads for hardware efficiency: CPU/GPU compute balance, memory management, data throughput, and networking
Develop monitoring and debugging tools for large-scale runs, enabling rapid diagnosis of performance regressions and failures
Deep experience in distributed systems, ML infrastructure, or high-performance computing (8+ years)
Production-grade expertise in Python
Low-level performance mastery: CUDA/cuDNN/Triton, CPU–GPU interactions, data movement, and kernel optimization
Scaling at the frontier: experience with PyTorch and training jobs using data, context, pipeline, and model parallelism
System-level mindset with a track record of tuning hardware–software interactions for maximum utilization
What You’ll Do
Design, build, and maintain large-scale data pipelines (batch and streaming) for robotics foundation model training and evaluation at petabyte scale
Own core data infrastructure: data model, storage systems, ingestion pipelines, transformation frameworks, and orchestration layers
Standardize data models and unify processing pipelines across real-world teleoperation and synthetic simulation datasets
Collaborate with a team of driven individuals committed to building general-purpose Physical AI
What You’ll Bring
Excellent software engineering skills (Python, Go, or similar)
Extensive experience designing, building, and maintaining large-scale data pipelines (8+ years)
Deep understanding of distributed systems (Spark, Kafka, or similar)
Extensive experience with data storage technologies (data lakes, warehouses, object stores like S3)
Experience running and maintaining production-grade infrastructure (Kubernetes, Terraform)
Bonus: Experience supporting AI systems, in particular embodied AI like self-driving
Robots will learn in simulation before they hit the factory. Genesis-World is our bet on that future.
Genesis-World is an open-source, general-purpose simulation platform for physical AI from Genesis AI. One unified multi-physics engine: rigid bodies, FEM, MPM, particles, cloth, fluids. A robot arm can pour water onto sand, grasp a deformable object, or cut a soft body, all in the same simulation. Nyx, our in-house renderer, may be the most promising renderer for robotics out there: real-time photo-realistic rendering, advanced features like depth of field, and state-of-the-art techniques never seen before. Sensors of every kind: cameras, lidar, IMU, contact forces, temperature, plus arguably the most advanced tactile simulation available (paper). And the engine keeps growing: we are developing internally the most comprehensive and fastest Incremental Potential Contact (paper) solver for deformable body dynamics we know of, soon to be open-sourced. It powers real business applications, from full-fledged box packaging with labelling machine and all, to wire harnessing and lab automation, without any physics hack or compromise.
Everything is Python-first and runs anywhere. Kernels are written once, and Quadrants, our in-house JIT compiler, lowers them to CUDA, AMD ROCm, Apple Metal, Vulkan, x86, and ARM64. A single laptop or a datacenter. Massively batched GPU simulation for learning at scale, and complex non-batched scenes where CPU wins outright.
This is at the core of Genesis AI's strategy. Evaluation is the bottleneck of scalable robotics: real hardware caps iteration at wall-clock time, but simulation turns it into a compute problem. Ours already runs two orders of magnitude faster than hardware (tens of thousands of episodes in half an hour instead of 200+ hours), while correlating with on-hardware rollouts at 89%. The north star: physical AI that improves at the speed of compute.
You push the physics of Genesis-World forward. The mandate is clear: ship production-ready simulation capabilities that matter for the company's internal needs. Research applied end-to-end, from algorithm to merged, tested, documented code that real robot-learning pipelines depend on. Occasional groundbreaking research happens, notably through academic collaborations. But the core of the job is making the engine measurably better along five axes:
Speed. Algorithms that are not only faster but also smart enough to spend compute only where it matters across both time and space: larger stable timesteps, selective fidelity (adaptive across scales or simply hand-set), structure-aware solvers.
Completeness. No physics off limits: water, human animation, air flow, gravel, tendons, even body organs. Whatever the next use-case needs, the engine grows to cover it.
Fidelity. More realistic models: contact, friction, deformation, energy, actuation, materials…
Versatility. Extensible multi-physics without compromise on realism: all solvers in the scene coupled together at once, two-way and constraint-based. Write your own solver and it joins the scene like a native one, growing into an open solver ecosystem.
Scalability. From workstation to factory scale, and one day, city scale: thousands of interacting entities, batched across environments, without losing physical soundness.
Our ambition is to establish Genesis-World as the go-to simulator for physical AI, from companies and research labs to individuals.
Every fidelity for every physics. The same physics at every point of the speed-accuracy spectrum, from heavily batched training with XPBD or VBD to final validation with IPC. Same scene, same API, pick your tradeoff.
Invent physics level-of-detail (LOD). Rendering has had LOD for decades, physics is still waiting. Simulate at full fidelity what agents interact with and see, coarsely what they do not.
Heterogeneous environments. Every parallel world can hold a completely different model: different bodies, joints, and collision geometries.
Adaptive timesteps per island. Error-based control with Runge-Kutta Dopri5, and Time-of-Impact stepping during collision detection, as done in Jiminy.
Couple everything, exactly. Efficient and accurate two-way constraint-based coupling between heterogeneous grey-box solvers, using state-of-the-art methods like ADMM. Owning every solver in the stack is what makes it possible.
More scalable constraint solvers. Push rigid constraint solving beyond its current scalability ceiling (reference).
Unify contact resolution. Hydro-elastic compliance, unilateral constraints, and sequential impulses in the same framework, ideally under one generic formulation.
Closed kinematic loops without constraints. Handle loops intrinsically for numerical stability and speed, in the spirit of Kamino.
Day to day: you write your physics in plain Python and Quadrants makes it fast on every backend. And you validate it the hard way: analytical closed forms, other engines, real-world data.
You are a physicist and an engineer at once. You judge a method by whether it holds up in production at real scale, and you do not stop until it does. No blind spots: you relentlessly hunt down even the defect that looks insignificant, because it never is.
A strong background in physics-based simulation, preferably related to robotics: RBD, FEM, MPM, SPH, IPC, XPBD, VBD, ABD, plus constrained optimization and numerical integration of stiff systems.
A track record of shipping simulation code that others rely on, in an engine, in industry, or in a research codebase used beyond its authors.
Solid HPC programming (CPU and/or GPU), and an instinct for what makes a numerical method fast in practice, beyond complexity classes.
Rigor in validation: analytical closed forms, cross-engine consistency, real-world data.
Bonus points: publications in simulation, graphics, or robotics venues (SIGGRAPH, ICRA, IROS, CoRL, RSS). Contributions to an open-source physics engine.
","descriptionPlain":"WHAT WE'RE BUILDING\n\nRobots will learn in simulation before they hit the factory. Genesis-World is our bet on that future.\n\nGenesis-World https://github.com/Genesis-Embodied-AI/genesis-world is an open-source, general-purpose simulation platform for physical AI from Genesis AI https://www.genesis.ai/. One unified multi-physics engine: rigid bodies, FEM, MPM, particles, cloth, fluids. A robot arm can pour water onto sand, grasp a deformable object, or cut a soft body, all in the same simulation. Nyx, our in-house renderer, may be the most promising renderer for robotics out there: real-time photo-realistic rendering, advanced features like depth of field, and state-of-the-art techniques never seen before. Sensors of every kind: cameras, lidar, IMU, contact forces, temperature, plus arguably the most advanced tactile simulation available (paper https://openreview.net/forum?id=md8q2kfZHP). And the engine keeps growing: we are developing internally the most comprehensive and fastest Incremental Potential Contact (paper https://ipc-sim.github.io) solver for deformable body dynamics we know of, soon to be open-sourced. It powers real business applications, from full-fledged box packaging with labelling machine and all, to wire harnessing and lab automation, without any physics hack or compromise.\n\nEverything is Python-first and runs anywhere. Kernels are written once, and Quadrants, our in-house JIT compiler, lowers them to CUDA, AMD ROCm, Apple Metal, Vulkan, x86, and ARM64. A single laptop or a datacenter. Massively batched GPU simulation for learning at scale, and complex non-batched scenes where CPU wins outright.\n\nThis is at the core of Genesis AI's strategy https://www.genesis.ai/blog/the-role-of-simulation-in-scalable-robotics-genesis-world-10-and-the-path-forward. Evaluation is the bottleneck of scalable robotics: real hardware caps iteration at wall-clock time, but simulation turns it into a compute problem. Ours already runs two orders of magnitude faster than hardware (tens of thousands of episodes in half an hour instead of 200+ hours), while correlating with on-hardware rollouts at 89%. The north star: physical AI that improves at the speed of compute.\n\n\n\n\nTHE ROLE\n\nYou push the physics of Genesis-World forward. The mandate is clear: ship production-ready simulation capabilities that matter for the company's internal needs. Research applied end-to-end, from algorithm to merged, tested, documented code that real robot-learning pipelines depend on. Occasional groundbreaking research happens, notably through academic collaborations. But the core of the job is making the engine measurably better along five axes:\n\n - Speed. Algorithms that are not only faster but also smart enough to spend compute only where it matters across both time and space: larger stable timesteps, selective fidelity (adaptive across scales or simply hand-set), structure-aware solvers.\n\n - Completeness. No physics off limits: water, human animation, air flow, gravel, tendons, even body organs. Whatever the next use-case needs, the engine grows to cover it.\n\n - Fidelity. More realistic models: contact, friction, deformation, energy, actuation, materials…\n\n - Versatility. Extensible multi-physics without compromise on realism: all solvers in the scene coupled together at once, two-way and constraint-based. Write your own solver and it joins the scene like a native one, growing into an open solver ecosystem.\n\n - Scalability. From workstation to factory scale, and one day, city scale: thousands of interacting entities, batched across environments, without losing physical soundness.\n\nOur ambition is to establish Genesis-World as the go-to simulator for physical AI, from companies and research labs to individuals.\n\n\n\n\nTHE PROBLEMS WAITING FOR YOU\n\n - Every fidelity for every physics. The same physics at every point of the speed-accuracy spectrum, from heavily batched training with XPBD or VBD to final validation with IPC. Same scene, same API, pick your tradeoff.\n\n - Invent physics level-of-detail (LOD). Rendering has had LOD for decades, physics is still waiting. Simulate at full fidelity what agents interact with and see, coarsely what they do not.\n\n - Heterogeneous environments. Every parallel world can hold a completely different model: different bodies, joints, and collision geometries.\n\n - Adaptive timesteps per island. Error-based control with Runge-Kutta Dopri5, and Time-of-Impact stepping during collision detection, as done in Jiminy.\n\n - Couple everything, exactly. Efficient and accurate two-way constraint-based coupling between heterogeneous grey-box solvers, using state-of-the-art methods like ADMM. Owning every solver in the stack is what makes it possible.\n\n - More scalable constraint solvers. Push rigid constraint solving beyond its current scalability ceiling (reference https://www.roboticsproceedings.org/rss20/p108.pdf).\n\n - Unify contact resolution. Hydro-elastic compliance, unilateral constraints, and sequential impulses in the same framework, ideally under one generic formulation.\n\n - Closed kinematic loops without constraints. Handle loops intrinsically for numerical stability and speed, in the spirit of Kamino https://disneyresearch.github.io/kamino.\n\nDay to day: you write your physics in plain Python and Quadrants makes it fast on every backend. And you validate it the hard way: analytical closed forms, other engines, real-world data.\n\n\n\n\nWHO YOU ARE\n\nYou are a physicist and an engineer at once. You judge a method by whether it holds up in production at real scale, and you do not stop until it does. No blind spots: you relentlessly hunt down even the defect that looks insignificant, because it never is.\n\n - A strong background in physics-based simulation, preferably related to robotics: RBD, FEM, MPM, SPH, IPC, XPBD, VBD, ABD, plus constrained optimization and numerical integration of stiff systems.\n\n - A track record of shipping simulation code that others rely on, in an engine, in industry, or in a research codebase used beyond its authors.\n\n - Solid HPC programming (CPU and/or GPU), and an instinct for what makes a numerical method fast in practice, beyond complexity classes.\n\n - Rigor in validation: analytical closed forms, cross-engine consistency, real-world data.\n\nBonus points: publications in simulation, graphics, or robotics venues (SIGGRAPH, ICRA, IROS, CoRL, RSS). Contributions to an open-source physics engine."},{"id":"e1af8522-b3fe-4887-8159-38e4e1523711","title":"Robotics","department":"Engineering & Research","team":"Robotics","employmentType":"FullTime","location":"London","secondaryLocations":[],"publishedAt":"2026-09-07T14:48:08.730+00:00","isListed":true,"isRemote":true,"workplaceType":"Hybrid","address":{"postalAddress":{"addressCountry":"United Kingdom","addressLocality":"London"}},"jobUrl":"https://jobs.ashbyhq.com/genesis/e1af8522-b3fe-4887-8159-38e4e1523711","applyUrl":"https://jobs.ashbyhq.com/genesis/e1af8522-b3fe-4887-8159-38e4e1523711/application","descriptionHtml":"Design, implement, and optimize the embedded control stack for general-purpose robots
Design motion planning and trajectory optimization algorithms for dynamic locomotion and manipulation
Build real-time state estimation pipelines for pose, contact, and force sensing, fusing heterogeneous sensor data under noise and uncertainty
Formulate and solve optimal control problems (nonlinear MPC, convex optimization, trajectory optimization) for high-performance and stable behavior
Build modular and robust software frameworks enabling rapid iteration between simulation and hardware
Lead debugging, tuning, and validation of controllers directly on physical robots
Passion for your craft and demonstrated excellence in control systems engineering
Extensive experience in designing, implementing, and deploying advanced control algorithms on real robotic system products (8+ years)
Deep expertise in dynamics, kinematics, optimal control, and state estimation
Strong command of hardware interfaces, sensors (IMUs, F/T sensors), and actuation technologies (motors, gearboxes, drivers)
Production-level mastery of C++ with a track record of building reliable, safety-critical software
Proven ability to bridge across hardware, software, and algorithms to deliver robust end-to-end systems
An open mind about the power of deep learning, reinforcement learning, and vision-language-action models as critical for general-purpose robotics
Bonus: Experience shipping humanoid robots or whole-body control systems
Bonus: Impactful published work in control theory, state estimation, or mathematical optimization
Bonus: Familiarity with parallel computation on GPUs to accelerate optimization
Build core simulation-centered infrastructure for robotics model development, evaluation, and iteration.
Design scalable systems for creating, running, and managing diverse simulation workloads across tasks, environments, and robots.
Develop reusable environment frameworks, task interfaces, and tooling that make simulation workflows more reliable and observable.
Work with the policy training team to bridge the sim-to-real gap
Develop a real-to-sim pipeline in collaboration with the asset generation team
Collaborate with a team of driven individuals committed to building general-purpose Physical AI
Passion for your craft and demonstrated excellence in physics simulation engineering
Exceptional ownership and initiative: finding and solving problems independently
Extensive experience in physics simulation, with a preference for expertise in rigid body simulation and FEM
Insights and requests from a robotic researcher/engineer’s perspective for improving user experience
Proficiency in CUDA and GPU programming
Built closed-loop or log-replay evaluation at scale.
Experience with robotics simulation tools such as genesis-world, Isaac Sim, MuJoCo, Newton, Drake, Gazebo, or similar platforms.
Experience with synthetic data, scene generation, asset pipelines, sim-to-real, or robotics benchmarks.
Experience with robot manipulation, mobile manipulation, humanoids, or embodied AI systems.
What You'll Do
Build machine learning models for robotics control end-to-end: data curation, careful evaluation, model architecture, training/inference stacks, rigorous experiments
Collaborate with simulation and real-world robotics teams to curate high-quality, diverse, and large-scale datasets
Curate the world’s best Internet-scale datasets for embodied perception and first-person robot video generation
Design new generative simulation techniques to expand simulation data scale and diversity, training and evaluating generative models of 3D objects and environments, and language/code models to generate tasks and reward functions
Collaborate with a team of driven individuals committed to building general-purpose Physical AI
What You'll Bring
Passion for your craft and demonstrated excellence in full-stack foundation model research and engineering
Exceptional ownership and initiative—finding and solving problems independently
Extensive experience pioneering new machine learning ideas or refining existing methods, supported by first-author publications or impactful projects (5+ years)
A relentless commitment to data and code quality, rigorous evaluation, and meticulous attention to detail
Production-level expertise in modern Python
Bonus: Experience with Vision-Language-Action models for robotics or web agents, embodied perception, or video generation
Design, build, and maintain foundational frameworks and tools to empower expert teams to experiment fast and turn ideas into full-fledged and production-ready systems.
Collaborate with expert teams to validate uses cases and build robust solutions, bringing cross-team perspective and aiming for modular and reusable components.
Identify and mitigate high-level code design flaws and development workflow inefficiencies that cause friction, hinder productivity, and create silos across the organization.
Advocate for good practices and maintain high code quality through code reviews, documentation, and training.
Excellent software engineering skills (e.g., in Python, Go, or Rust).
Extensive experience designing, building, and maintaining modular and reusable systems.
Experience collaborating with engineers and researchers of any fields, understanding complex requirements, and providing clear and simple solutions.
Ability to implement large-impact changes across the organization with minimal disruption.
Experience building internal tooling to accelerate AI research and development workflows (e.g., debugging, evaluation, experiment tracking and versioning, CI automation, etc.).
We are looking for a Technical Program Manager to work directly with our Head of Deployment to drive execution across our most critical customer deployments. You will operate at the center of complex engineering programs to coordinate AI, robotics, hardware, operations, and customer teams to move our systems from prototype to real-world production.
Partner directly with the Head of Deployment to drive end-to-end execution of customer deployments.
Own cross-functional programs spanning robotics software, AI, hardware, infrastructure, operations, and customer success.
Coordinate engineering milestones across multiple teams and ensure critical dependencies stay on track.
Drive site readiness, including integration planning, testing, validation, site logistics, and launch execution.
Engage directly with enterprise customers to translate technical requirements, report progress, and manage timelines.
Identify execution risks early and work across engineering teams to resolve technical and operational blockers.
Build scalable processes for deployment planning, release management, and customer onboarding.
Define how Genesis deploys and scales physical AI and robotics systems across diverse customer environments.
5+ years of experience in technical program management, systems engineering, or product operations for complex hardware/software platforms.
Technical depth to comfortably navigate technical discussions with software engineers, roboticists, and ML researchers.
Track record of leading cross-functional engineering programs from early planning through production rollout.
Clear written and verbal communication skills to align distributed engineering teams, leadership, and external enterprise stakeholders.
Strong systems thinking with a habit of bringing structure to ambiguous operational problems.
Willingness to travel to customer sites as needed to support critical deployment milestones.
Bonus: Direct experience deploying robotics or autonomous systems into real-world environments.
Bonus: Technical background in autonomy, perception, controls, simulation, machine learning, or field engineering.
Quadrants is an open-source translation layer that turns ordinary Python into optimized machine code at runtime, across CPU (Arm64, x86) and GPU (AMDGPU, CUDA, Metal, Vulkan) backends. CUDA is the primary target, with x86 the strong runner-up, especially for real-time simulation. The others drive adoption and support fast prototyping for research.
The ambition is explicit: build the best “Python compiler” in the world for physics and performance-critical numerical computing. We are fully committed to it, and we intend to beat leading alternatives like Nvidia Warp, Numba, or Triton.
It’s a standalone project with its own roadmap, built in the open. Genesis, our physics simulator, is its first and top-priority user, the source of its hardest workloads and its benchmarks. But Quadrants is general-purpose infrastructure for anyone writing performance-critical numerical Python, not an internal tool that happens to be public.
The bet: high-performance code should feel natural to write. Engineers write plain Python in a NumPy or PyTorch style, and Quadrants makes it fast.
You own Quadrants end to end: the API, the IR and lowering, the optimization passes, the numerical primitives, and the roadmap.
Our ultimate objective: engineers write for readability and maintainability, never for performance. They write the version of an algorithm they are most comfortable with, and Quadrants rebuilds it into kernels that run as if an expert had hand-tuned them. Performance is the compiler’s job, not theirs. Getting there is a frontier problem, part research and part engineering, and we intend to define the state of the art on it, relentlessly pushing the limit further.
It calls for two things at once: the low-level GPU and compiler skill to make a single kernel optimal, and the ability to keep the whole system coherent as it grows.
The core tradeoff is settled: usability comes first. When usability and performance collide, usability wins, unless the cost is severe. We will limit what the system supports to go fast (static shapes and fixed-size allocation baked into kernels, say), but never the convenience of the API itself. Experts who want maximum performance and full control over the generated code get lower-level escape hatches, down to inline assembly, C, or PTX.
Your mandate is to build the best open-source Python compiler for physics simulation out there. No limit on what that takes. If it means replacing the entire engine, your call. Either way, you keep Quadrants healthy and unblocking the company in the meantime.
API & authoring experience. The surface that physics and numerical-computing engineers write against, and, as an open-source library, the product’s face to the world. Idiomatic, natural to Python users, never a separate language bolted onto Python. Clean decorators, legible constraints, sharp errors, real escape hatches.
IR, lowering & backends. A stable IR that lowers to multiple backends, with room to grow and the door kept open for differentiable simulation. Today the computation graph is expressed through Quadrants’ own constructs. Recovering it automatically from ordinary Python is hard, and a long-term goal rather than today’s reality.
Optimization. Turn naive code into strong execution plans: vectorization, register and shared-memory usage, barriers and atomics, cooperative threading, dead-buffer and redundant-traffic elimination, dropping needless synchronization. Move low-level optimization off users and into the platform.
Primitives. Cholesky, eigendecomposition, matmul, reductions, solver kernels. Each owned with its correctness, benchmarks, and regression tracking.
Watching Genesis and the compiler together. Hunt for performance wins on both sides, in Genesis and in Quadrants itself. Propose API changes that cut duplication, improve readability, and reduce platform-specific special-casing. Less to maintain, more that simply works.
The project as a product. Benchmark real workloads, track performance over time, catch regressions early. Stay close to users (Genesis first), find friction and missing abstractions, drive adoption, set direction.
You reason about whole systems and have taste for where complexity should live. You measure before you optimize, and you reach for memory traffic and launch overhead before FLOPs. You have opinions about API design. You write tests and benchmarks as you go, not afterward.
The ground that judgment stands on:
GPU. What optimal looks like and why: access patterns, registers, shared memory, barriers, atomics, threading. CPU intuition a plus.
Python & API taste. Idiomatic Python. A feel for what’s natural to Python users. APIs that free users rather than constrain them.
Compilers. The path from authoring to kernels (IR, lowering, optimization), weighed against maintainability.
And how you work: proactive and largely self-directed. You fetch context across the company instead of waiting for it. You’re comfortable talking to the engineers who use what you build, and at ease designing in the open for users you’ll never meet. Ready to grow into leading a small team (a couple of people, near term).
People adopt Quadrants from outside Genesis. Engineers write clearer code with less boilerplate. Kernels keep getting faster. Performance is tracked, not guessed. The architecture gets easier to extend, not harder.
How you work is up to you. What matters is that your ownership grows over time: blind spots shrink, the parts you can vouch for expand, technical debt goes down, the core stays sound.
That takes treating understanding as real work, not overhead. You read and challenge the codebase instead of just stacking features on it. You take nothing you did not write for granted. You watch the competition and what’s coming next, and you bring it in and ship it.
Use AI agents if they help. They are productivity tools, not code owners. You are.
","descriptionPlain":"WHAT QUADRANTS IS\n\nQuadrants is an open-source translation layer that turns ordinary Python into optimized machine code at runtime, across CPU (Arm64, x86) and GPU (AMDGPU, CUDA, Metal, Vulkan) backends. CUDA is the primary target, with x86 the strong runner-up, especially for real-time simulation. The others drive adoption and support fast prototyping for research.\n\nThe ambition is explicit: build the best “Python compiler” in the world for physics and performance-critical numerical computing. We are fully committed to it, and we intend to beat leading alternatives like Nvidia Warp, Numba, or Triton.\n\nIt’s a standalone project with its own roadmap, built in the open. Genesis, our physics simulator, is its first and top-priority user, the source of its hardest workloads and its benchmarks. But Quadrants is general-purpose infrastructure for anyone writing performance-critical numerical Python, not an internal tool that happens to be public.\n\nThe bet: high-performance code should feel natural to write. Engineers write plain Python in a NumPy or PyTorch style, and Quadrants makes it fast.\n\n\n\n\nTHE ROLE\n\nYou own Quadrants end to end: the API, the IR and lowering, the optimization passes, the numerical primitives, and the roadmap.\n\nOur ultimate objective: engineers write for readability and maintainability, never for performance. They write the version of an algorithm they are most comfortable with, and Quadrants rebuilds it into kernels that run as if an expert had hand-tuned them. Performance is the compiler’s job, not theirs. Getting there is a frontier problem, part research and part engineering, and we intend to define the state of the art on it, relentlessly pushing the limit further.\n\nIt calls for two things at once: the low-level GPU and compiler skill to make a single kernel optimal, and the ability to keep the whole system coherent as it grows.\n\nThe core tradeoff is settled: usability comes first. When usability and performance collide, usability wins, unless the cost is severe. We will limit what the system supports to go fast (static shapes and fixed-size allocation baked into kernels, say), but never the convenience of the API itself. Experts who want maximum performance and full control over the generated code get lower-level escape hatches, down to inline assembly, C, or PTX.\n\nYour mandate is to build the best open-source Python compiler for physics simulation out there. No limit on what that takes. If it means replacing the entire engine, your call. Either way, you keep Quadrants healthy and unblocking the company in the meantime.\n\n\n\n\nWHAT YOU’LL OWN\n\n - API & authoring experience. The surface that physics and numerical-computing engineers write against, and, as an open-source library, the product’s face to the world. Idiomatic, natural to Python users, never a separate language bolted onto Python. Clean decorators, legible constraints, sharp errors, real escape hatches.\n\n - IR, lowering & backends. A stable IR that lowers to multiple backends, with room to grow and the door kept open for differentiable simulation. Today the computation graph is expressed through Quadrants’ own constructs. Recovering it automatically from ordinary Python is hard, and a long-term goal rather than today’s reality.\n\n - Optimization. Turn naive code into strong execution plans: vectorization, register and shared-memory usage, barriers and atomics, cooperative threading, dead-buffer and redundant-traffic elimination, dropping needless synchronization. Move low-level optimization off users and into the platform.\n\n - Primitives. Cholesky, eigendecomposition, matmul, reductions, solver kernels. Each owned with its correctness, benchmarks, and regression tracking.\n\n - Watching Genesis and the compiler together. Hunt for performance wins on both sides, in Genesis and in Quadrants itself. Propose API changes that cut duplication, improve readability, and reduce platform-specific special-casing. Less to maintain, more that simply works.\n\n - The project as a product. Benchmark real workloads, track performance over time, catch regressions early. Stay close to users (Genesis first), find friction and missing abstractions, drive adoption, set direction.\n\n\n\n\nWHO YOU ARE\n\nYou reason about whole systems and have taste for where complexity should live. You measure before you optimize, and you reach for memory traffic and launch overhead before FLOPs. You have opinions about API design. You write tests and benchmarks as you go, not afterward.\n\nThe ground that judgment stands on:\n\n - GPU. What optimal looks like and why: access patterns, registers, shared memory, barriers, atomics, threading. CPU intuition a plus.\n\n - Python & API taste. Idiomatic Python. A feel for what’s natural to Python users. APIs that free users rather than constrain them.\n\n - Compilers. The path from authoring to kernels (IR, lowering, optimization), weighed against maintainability.\n\nAnd how you work: proactive and largely self-directed. You fetch context across the company instead of waiting for it. You’re comfortable talking to the engineers who use what you build, and at ease designing in the open for users you’ll never meet. Ready to grow into leading a small team (a couple of people, near term).\n\n\n\n\nWHAT GREAT LOOKS LIKE\n\nPeople adopt Quadrants from outside Genesis. Engineers write clearer code with less boilerplate. Kernels keep getting faster. Performance is tracked, not guessed. The architecture gets easier to extend, not harder.\n\n\n\n\nWHAT WE’RE UNCOMPROMISING ABOUT\n\nHow you work is up to you. What matters is that your ownership grows over time: blind spots shrink, the parts you can vouch for expand, technical debt goes down, the core stays sound.\n\nThat takes treating understanding as real work, not overhead. You read and challenge the codebase instead of just stacking features on it. You take nothing you did not write for granted. You watch the competition and what’s coming next, and you bring it in and ship it.\n\nUse AI agents if they help. They are productivity tools, not code owners. You are."},{"id":"e1c4dcea-4c49-44cf-8594-3103289c2638","title":"Robot Learning","department":"Engineering & Research","team":"Robotics","employmentType":"FullTime","location":"London","secondaryLocations":[],"publishedAt":"2026-09-07T15:34:42.325+00:00","isListed":true,"isRemote":true,"workplaceType":"Hybrid","address":{"postalAddress":{"addressCountry":"United Kingdom","addressLocality":"London"}},"jobUrl":"https://jobs.ashbyhq.com/genesis/e1c4dcea-4c49-44cf-8594-3103289c2638","applyUrl":"https://jobs.ashbyhq.com/genesis/e1c4dcea-4c49-44cf-8594-3103289c2638/application","descriptionHtml":"What You’ll Do
Develop and optimize a learning-based robotic manipulation control stack
Design and maintain a teleoperation system with smooth, precise motion and low latency
Train robotic policies for manipulation and locomotion with reinforcement learning and imitation learning
Deploy robotic policies and diagnose latency or bottlenecks in the control pipeline
Analyze and minimize the sim-to-real gap by co-optimizing simulation and real-world robot behavior
Collaborate with a team of driven individuals committed to building general-purpose Physical AI
What You’ll Bring
Passion for your craft and demonstrated excellence in robotics engineering
Exceptional ownership and initiative—finding and solving problems independently
Focus, attention to detail, patience, and a methodical approach to complex tasks
Production-level expertise in modern Python or C++
Extensive experience building and deploying real-world learning-based robotic systems (5+ years)
Bonus: Hands-on experience troubleshooting and maintaining robotic systems across mechanical, electrical, and software components
About Genesis AI
Genesis AI builds the simulation platform powering the next generation of robotics and embodied AI. Our flagship product, Genesis World, is a unified multi-physics simulation engine paired with Nyx, our in-house GPU-accelerated, path-traced photorealistic renderer. Together they provide a platform for training robotics and embodied AI at scale; letting teams simulate complex physical worlds with high fidelity and speed, from a laptop to datacenter-grade GPUs, and close the gap between simulation and the real world.
About the Role
A simulator is only as useful as the content that runs inside it. Training embodied agents at scale means large volumes of assets that are simultaneously physically valid and visually convincing and that is a pipeline problem before it is an art problem.
We're hiring a Senior Engineer to build the path from raw input to sim-ready asset: geometry from a range of sources going in; clean meshes, correct collision proxies, plausible physical properties, materials that hold up under a path tracer, and reliable metadata coming out.
The bar is throughput and repeatability. An asset that needs a human pass to be usable doesn't scale, so the pipeline has to make good decisions on its own and flag the cases where it can't.
Part of that autonomy comes from agents embedded in the graph itself. Nodes can call models to produce input content, drive generation, pick parameters, interpret a mesh before deciding how to treat it, or judge whether a result is acceptable. Making that reliable is its own engineering problem: agents are non-deterministic, they fail in ways a mesh operator doesn't, and they need guardrails, structured outputs, retries and verification around them. You'll be building that as a first-class part of the pipeline rather than bolting it on.
You'll work closely with the stakeholders, who set direction and priorities. You own the technical design and the implementation and you're expected to bring enough pipeline experience to challenge the plan where it needs challenging.
What You'll Do
Build and maintain the end-to-end asset pipeline, from ingestion through to validated output
Design agent-driven nodes that generate input content and make decisions inside the graph, with the structured outputs, guardrails and fallbacks that make them dependable in batch
Develop automated processing stages: mesh repair, retopology, decimation and LODs, UV unwrapping and texel density normalization, collision proxy generation
Derive physical properties so assets behave correctly in simulation: mass, inertia, contact parameters, articulation setup
Handle appearance: material conversion to PBR, texture processing, baking, and color management so assets render correctly in Nyx
Define and enforce asset standards (units, orientation, pivots, naming, provenance) and build the validators that reject noncompliant assets before they reach the library
Scale the pipeline to large batch runs, with caching, versioning and selective reprocessing when a stage changes
Build quality checks into the pipeline itself: physics stability harnesses, render smoke tests, automated evaluation of generated content, comparison on reprocessed assets
Work with the rendering, physics and research teams on what sim-ready needs to mean for each downstream use
What We're Looking For (ideally)
5+ years building content or asset pipelines in games, VFX, simulation, or 3D tooling, with ownership of a pipeline in production
Deep knowledge of 3D asset formats and their quirks: USD, glTF, FBX, OBJ, plus physics description formats such as URDF or MJCF
Strong Python and C++ where processing needs to be fast
Hands-on mesh processing expertise: remeshing, decimation, UV generation, repair operations, convex decomposition, and a clear sense of how each one fails
Solid command of PBR material authoring, texture baking and color spaces
Experience driving DCC tools headlessly (Blender, Houdini, Maya) and building batch or distributed job systems around them
Practical experience putting LLM or agent calls into production systems: prompting, tool use, structured output, cost and latency control, and handling non-determinism
Understanding of what makes an asset physically valid in a simulator: visual versus collision geometry, mass properties, contact behavior, numerical stability
Judgment about quality bars, including when an asset is good enough and when it isn't
Nice to Have
Experience processing output from generative or reconstruction-based 3D sources, and evaluating that output automatically
Agent frameworks, orchestration and eval tooling applied to content generation
OpenUSD at scale: composition, layering, MaterialX
Semantic annotation and dataset curation at volume
Exposure to simulation stacks such as MuJoCo, Isaac Sim or similar
What You'll Do
Own and scale the company's internal IT infrastructure across identity, devices, productivity systems, networking, and business applications
Design and implement identity and access management systems, including SSO, MFA, lifecycle management, and role-based access controls across platforms such as Okta, Google Workspace, Slack, GitHub, and other internal tools
Lead company-wide IT initiatives including domain migrations, email infrastructure, device management, application integrations, and employee onboarding/offboarding workflows
Build and automate IT operations through scripting, workflow automation, and infrastructure tooling
Manage endpoint infrastructure across macOS, Linux, and Windows environments, including device provisioning, compliance, and lifecycle management
Partner closely with engineering, operations, and business teams to deploy and support internal systems that enable company growth and productivity
Evaluate, implement, and manage business-critical SaaS platforms and vendor relationships
Establish scalable IT policies, processes, and documentation to support a rapidly growing organization
Maintain foundational security controls across identity, endpoints, networks, and corporate systems
Support security initiatives including access reviews, incident response, vulnerability remediation, and security awareness programs
Troubleshoot and resolve complex IT issues across hardware, software, networking, and cloud-based systems
Collaborate with a team of driven individuals committed to building general-purpose Physical AI
What You'll Bring
Strong experience building and operating IT infrastructure in fast-growing startups or technical organizations
Hands-on experience with identity and access management platforms such as Okta, Google Workspace, Microsoft Entra ID, or similar systems
Experience leading domain migrations, email administration, SSO deployments, and SaaS application integrations
Strong scripting and automation skills using Python, Bash, Go, or similar languages
Experience managing endpoint fleets through MDM platforms such as Kandji, Jamf, Intune, or similar tools
Strong understanding of networking fundamentals, corporate infrastructure, and cloud environments
Experience administering and integrating modern workplace tools such as Slack, GitHub, Notion, Atlassian products, Google Workspace, and productivity platforms
Working knowledge of security best practices across identity management, endpoint security, access controls, and infrastructure operations
Ability to operate independently, prioritize effectively, and build systems from zero to scale
Strong troubleshooting and problem-solving skills across hardware, software, networking, and SaaS environments
Excellent communication skills and ability to support both technical and non-technical stakeholders
Bonus: Experience supporting engineering-heavy organizations, AI companies, robotics companies, or high-growth startups
Bonus: Experience with compliance frameworks such as SOC 2, ISO 27001, or similar security and operational standards
We are a full-stack general-purpose robot company, born across San Francisco, London and Paris and bound by a single mission: to make general-purpose robots a reality, unlocking infinite physical labor for the world and freeing people for creativity, exploration and the pursuits they love for their own sake.
We're building the engine that will teach robots to do real work in the real world. We recently unveiled Eno, our wheeled, dexterous robot powered by our GENE foundation model and we are starting to put it to work with real customers.
General-purpose robots learn from real, dexterous human work, the kind of data that does not exist on the internet. We capture it with our own collection hardware: a person wears it, does the work, and the robot learns to do it too.
You will be a senior member of the team scaling our data collection operation in Paris, working closely with our Paris research and engineering teams and with the operation in the Bay Area.
You will help lead the day-to-day execution and quality bar on the Paris floor while remaining deeply hands-on with data collection, teleoperation, task setup, and real-robot testing. You will wear our gloves and cameras, teleoperate our hardware, and take ownership of complex collection requests from setup through execution and quality review.
You will work shoulder to shoulder with our model engineering and research teams, turning their requests into physical demonstrations and feeding what you learn on the floor back to them. You will be expected to operate autonomously, help guide and train other operators and contribute to how the Paris operation evolves as the team grows.
This is a hands-on leadership role. You will not just run existing tasks: you will help shape how they are executed, how quality is measured, and how other operators are trained to deliver consistently.
Help lead the Paris data collection floor. Set the execution and quality bar for day-to-day collection work and help the team scale effectively.
Own complex collection setups. Take a loosely defined request and turn it into a working, repeatable collection task, including the setup, fixtures, props, environment, and procedure, with minimal support from engineering.
Create high-quality training data. Wear our gloves and cameras, teleoperate our hardware, and produce precise demonstrations that can be used to train our models.
Own the full collection workflow. Go beyond collecting the data itself: review and quality-check your work, identify unusable episodes, support annotation, and help ensure that what reaches research is trainable data.
Build repeatable tasks. Turn new collection requests into clear, repeatable processes that can be executed consistently by other operators.
Partner with research and model engineering. Translate their requests into physical collection tasks, share observations from the floor, and iterate with them on setups and collection strategy.
Improve the operation. Identify when a fixture, station layout, procedure, or collection method can be improved and proactively help implement those improvements.
Write and improve the playbook. Own SOPs, quality checks, and training materials that allow the Paris operation to scale while maintaining a consistent quality bar.
Guide other operators. Help onboard, train, and mentor operators and reinforce consistent execution across the team.
Run real-robot testing. Partner with reliability and engineering teams to test robots on the floor, document what you observe, and help surface issues.
Own quality. Review your own work and the team's, take direct feedback, and keep raising the bar.
You have done precise, repetitive physical work to a high quality bar and can explain what made your work good and how you measured it. This could include data collection, teleoperation, motion capture, robotics lab work, assembly, fine craft, or another hands-on technical environment.
Hands-on robot or glove-based data collection experience is a strong plus, but not a requirement. This kind of data collection is new, and we will teach you our specific tools and approach.
You build, not just run. Give you a vague request and a scrappy setup, and you can turn it into a clean, repeatable task with the right fixtures, environment, process, and written procedure.
You can operate autonomously. You are comfortable taking ownership of complex collection tasks with limited day-to-day support from engineering.
You understand quality beyond collection. You can review your work, recognize when collected data is unusable or below the quality bar, and understand why.
You are ready to lead. You have trained, mentored, or led others doing hands-on work and want to help establish the execution and quality standard for a growing team.
You work well with research and engineering teams. You can translate their requests into physical collection tasks and turn what you observe during collection into useful feedback.
You have exceptional attention to detail. You notice the moment something about the setup, hardware, environment, procedure, or collected data is off.
You are technically handy. You are comfortable with Linux, some Python scripting to automate repetitive work, and hardware assembly and troubleshooting.
You are comfortable in a fast-moving environment. Hardware, processes, and collection methods will evolve quickly, and you are comfortable adapting while helping improve how the team works.
You communicate well in English. You will work daily with teams in Paris and the Bay Area across time zones. French is a plus.
Bonus: reliability testing. You have partnered with a reliability or test team on real-robot testing.
Work takes place in production environments and technical laboratories.
Prolonged standing and repetitive hand and arm movements while wearing collection equipment, including gloves, head-mounted cameras, and headsets, for extended periods.
Handling and lifting of loads, including props, fixtures, robot equipment, and station equipment, up to 23 kg for station setup and reconfiguration.
Real ownership. You will have meaningful ownership over how the Paris data collection floor operates and help set the standard the team grows toward.
Hands-on leadership. You will remain close to the work while helping guide other operators, improve execution, and raise the quality bar.
Build, not just execute. You will take ambiguous collection needs and turn them into reliable, repeatable tasks that can scale across the team.
A clear growth path. As the operation scales, this role can grow toward leading or supervising teams of operators or toward broader project and program management responsibilities.
The front line of the mission. The data you create, review, and improve is what teaches our robots to work.
About Genesis
Genesis is a physical AI research lab and full-stack robotics company building a universal robotics foundation model (one AI \"brain\" that can power many kinds of robots) and a horizontal robotics platform. We work in San Francisco, Paris and London and combine large-scale AI research, an in-house physics simulation engine, and real robot hardware under one roof.
The role
You'll be the person who turns our most ambitious bets into shipped results. You'll sit between researchers, hardware engineers, software engineers and leadership, across two continents, making sure everyone knows what matters, who owns it and when it lands.
What you'll do
Drive cross-functional programs at the frontier of AI and robotics, from foundation model training to hardware–software integration, with pace and accountability
Turn ambitious goals into lean, unambiguous roadmaps: clear deliverables, clear owners, clear timelines
Dive into robotics hardware, software, simulation, and large-scale AI to understand trade-offs, surface uncertainties, and ask the questions others miss
Keep Bay Area and Paris teams tightly aligned, dependencies unblocked, and risks visible early
Build lightweight systems (roadmaps, metrics, dashboards) that keep execution focused on what matters
Leverage AI to build automated program management workflows, so status updates, risk tracking, and reporting run themselves and the team spends its time on real decisions
What you'll bring
4+ years leading complex technical programs end-to-end in fast-moving environments (startup, research lab, or high-growth tech)
Enough technical fluency to hold a real conversation with an ML researcher about a training run and with a hardware engineer about an integration test, so you can challenge an estimate, spot a hidden dependency, or flag a risk before it becomes a delay
A track record of learning a new technical domain quickly and becoming useful in it within weeks, not months
Crisp, direct communication that aligns researchers, engineers, and leadership
Bias for action: many decisions can be reversed and you favor speed over endless analysis
Ownership: you own the outcome and drive the process to get tangible results
Bonus
Experience in robotics, hardware/software integration, or research workflows
Fluency in French
Genesis AI is building a universal robotics foundation model — the \"brain\" that lets robots see, reason and handle objects in the real world. Our models learn from large volumes of robot, simulation and human-demonstration data, and the quality of that data directly shapes how well our robots perform.
We're scaling our labeling team. You'll review and annotate robot and video data that trains and evaluates our models, working side by side with the researchers who use it. Your judgment on hard cases will shape our guidelines, our tools and, ultimately, how our robots behave.
Annotate robot episodes, videos, images, and sensor data using internal labeling tools
Label objects, actions, task steps, and successes or failures in robot demonstrations
Write clear, consistent natural-language descriptions of robot and human behavior
Flag ambiguous or edge cases and help define how they should be handled
Perform quality assurance checks on teammates' annotations and join regular calibration sessions
Meet quality and throughput targets while tracking your own accuracy over time
Report tool bugs, workflow friction, and guideline gaps to the engineering team
2-5+ years of data annotation or labeling experience in a detail-heavy, deadline-driven environment
A strong sense of ownership: you see work through to the finish line
Drive and urgency: you're hungry to learn, move fast, and keep going on tight deadlines
A track record of high-accuracy work at volume, with the ability to hold quality while speed increases
Sharp attention to detail and the judgment to spot when instructions fall short
Excellent written communication, and the confidence to raise problems early
Comfort with ambiguity and shifting priorities in a fast-moving startup
Curiosity, coach-ability, and consistency
Prior data annotation, quality assurance, or issue triage experience
Experience with video, 3D, point-cloud, text, or robotics data
Interest in robotics, machine learning, or how physical tasks are performed
Genesis AI is an equal opportunity employer. We welcome applicants of all backgrounds and do not discriminate on the basis of race, religion, color, national origin, sex, sexual orientation, gender identity, age, disability or veteran status. Pay range reflects the expected hourly rate for this role in the San Francisco Bay Area; final rate depends on skills and experience.
","descriptionPlain":"ABOUT GENESIS AI\n\nGenesis AI is building a universal robotics foundation model — the \"brain\" that lets robots see, reason and handle objects in the real world. Our models learn from large volumes of robot, simulation and human-demonstration data, and the quality of that data directly shapes how well our robots perform.\n\n\n\n\nTHE ROLE\n\nWe're scaling our labeling team. You'll review and annotate robot and video data that trains and evaluates our models, working side by side with the researchers who use it. Your judgment on hard cases will shape our guidelines, our tools and, ultimately, how our robots behave.\n\n\n\n\nWHAT YOU'LL DO\n\n - Annotate robot episodes, videos, images, and sensor data using internal labeling tools\n\n - Label objects, actions, task steps, and successes or failures in robot demonstrations\n\n - Write clear, consistent natural-language descriptions of robot and human behavior\n\n - Flag ambiguous or edge cases and help define how they should be handled\n\n - Perform quality assurance checks on teammates' annotations and join regular calibration sessions\n\n - Meet quality and throughput targets while tracking your own accuracy over time\n\n - Report tool bugs, workflow friction, and guideline gaps to the engineering team\n\n\n\n\nWHAT WE'RE LOOKING FOR\n\n - 2-5+ years of data annotation or labeling experience in a detail-heavy, deadline-driven environment\n\n - A strong sense of ownership: you see work through to the finish line\n\n - Drive and urgency: you're hungry to learn, move fast, and keep going on tight deadlines\n\n - A track record of high-accuracy work at volume, with the ability to hold quality while speed increases\n\n - Sharp attention to detail and the judgment to spot when instructions fall short\n\n - Excellent written communication, and the confidence to raise problems early\n\n - Comfort with ambiguity and shifting priorities in a fast-moving startup\n\n - Curiosity, coach-ability, and consistency\n\n\n\n\nBONUS QUALIFICATIONS\n\n - Prior data annotation, quality assurance, or issue triage experience\n\nExperience with video, 3D, point-cloud, text, or robotics data\n\n - Interest in robotics, machine learning, or how physical tasks are performed\n\n\n\nGenesis AI is an equal opportunity employer. We welcome applicants of all backgrounds and do not discriminate on the basis of race, religion, color, national origin, sex, sexual orientation, gender identity, age, disability or veteran status. Pay range reflects the expected hourly rate for this role in the San Francisco Bay Area; final rate depends on skills and experience."},{"id":"096716ee-52be-4f4b-90bf-22cd509c415b","title":"Senior Manager, Supply Chain (Robot Production)","department":"Manufacturing","team":"Manufacturing","employmentType":"FullTime","location":"Bay Area","secondaryLocations":[],"publishedAt":"2026-10-01T22:12:16.810+00:00","isListed":true,"isRemote":false,"workplaceType":"OnSite","address":{"postalAddress":{"addressRegion":"California","addressCountry":"United States","addressLocality":"San Carlos"}},"jobUrl":"https://jobs.ashbyhq.com/genesis/096716ee-52be-4f4b-90bf-22cd509c415b","applyUrl":"https://jobs.ashbyhq.com/genesis/096716ee-52be-4f4b-90bf-22cd509c415b/application","descriptionHtml":"We are a full-stack general-purpose robot company, born across San Francisco and Paris and bound by a single mission: to make general-purpose robots a reality, unlocking infinite physical labor for the world and freeing people for creativity, exploration, and the pursuits they love for their own sake.
We are French and American by birth and global by nature. We are backed by partners who share our vision, including Eclipse, Khosla Ventures, Bpifrance, and HSG, alongside Eric Schmidt, Xavier Niel, Daniela Rus, and Vladlen Koltun. Our people work across the Bay Area and Europe, building the engine that will teach robots to do real work in the real world. We recently unveiled Eno, our wheeled, dexterous robot powered by our GENE foundation model, and we are starting to put it to work with real customers.
Every part that goes into our robot: who makes it, at what price, on what lead time, to what quality. This is the hands-on role. You build the supplier base for production, develop domestic sources for the sub-systems that need them, and make sure the line never waits on a part. You start with a bill of materials, a design team and a few suppliers; you end with a qualified, dual-sourced supply chain that supports production with our manufacturing partner. You are at our facility, at the partner and at suppliers more than you are at a desk.
Be on the floor: at our facility, at our manufacturing partner and at suppliers; first articles, line-down calls and recoveries are yours.
Build the supplier base: source, qualify and manage suppliers for machined and cast structures, gears and actuators, motors, PCBAs, harnesses, shells, batteries and sensors; two sources for every critical sub-system.
Develop domestic suppliers for the sub-systems that need them, and keep the content records that come with selling into regulated markets.
Run the commercial work day to day: RFQs, should-cost, piece price and tooling negotiation, payment and warranty terms; work with Legal on supplier agreements.
Sit in design reviews from the start: flag long-lead, single-source and high-risk parts before design freeze; bring supplier capability into the design.
Own clear-to-build and the plan: every unit on the schedule has a complete kit before it starts; a rolling plan the line, the partner and the suppliers all see; capacity planning with suppliers ahead of demand; inventory accuracy and safety stock by part class.
Own the BOM in production: cost roll-ups by phase, engineering change control with the design team and Manufacturing Engineering, cost-down from pilot to volume.
Build the supplier quality loop with Manufacturing Engineering and Quality: first article and PPAP where it applies, incoming inspection, supplier scorecards and quarterly reviews, corrective action and recovery.
Run logistics: inbound, between our facility and our partner's, import and export paperwork with Legal, spares and service parts.
Run the systems we have and help stand up the next ones: part numbering, BOM control, MRP; own the weekly numbers (clear-to-build, on-time delivery, PPV, inventory).
You have built a supplier base for a hardware product going into production, not managed or handed to you, and you produce results in weeks, not quarters.
Deep experience with machined and cast metal parts, gears or actuators, motors, PCBAs and harnesses, and the suppliers who make them; you read drawings and GD&T.
You have developed new suppliers in new regions and qualified them to a standard.
Strong commercial record: you have negotiated price, tooling and terms with suppliers and contract manufacturers.
You are fluent in BOMs, change control, should-cost and cost roll-ups, and comfortable with the people who design the product.
8 or more years in supply chain, sourcing or supplier development for complex hardware, with time in a mature organization so you know what the systems look like at scale; you have led a small team or are ready to; engineering degree preferred.
Bonus: robotics or automotive; contract manufacturer selection and transition; import and export compliance; APICS or CPSM.
Office and production environments; travel to suppliers and partner sites, including international, up to 30%.
We are always looking for people who are driven to work on hard problems and want to shape what the world will look like in ten years. A few of the values that guide us:
Limitless ambition. We hold ourselves to the highest bar in everything we build.
Relentless urgency. Iterate faster, learn faster. Time is the denominator.
Type-2 Fun. Take the work seriously, not yourself. Embrace the grind and find joy in the struggle.
Radical transparency. Candid communication builds trust, and direct feedback drives growth.
Multiculturalism. We come from across the world. Be curious. Meet people on their terms.
Team over ego. Help others, share context, and win as a team.
A joyful journey. We take pride in the craft, enjoy the ride, and make space to laugh. Happiness is part of how we get there.
We are a full-stack general-purpose robot company, born across San Francisco and Paris and bound by a single mission: to make general-purpose robots a reality, unlocking infinite physical labor for the world and freeing people for creativity, exploration, and the pursuits they love for their own sake.
We are French and American by birth and global by nature. We are backed by partners who share our vision, including Eclipse, Khosla Ventures, Bpifrance, and HSG, alongside Eric Schmidt, Xavier Niel, Daniela Rus, and Vladlen Koltun. Our people work across the Bay Area and Europe, building the engine that will teach robots to do real work in the real world. We recently unveiled Eno, our wheeled, dexterous robot powered by our GENE foundation model, and we are starting to put it to work with real customers.
We run two kinds of robot hardware. On our data collection floor, a growing number of robot stations and wearable collection kits run all day, every day, and the floor is scaling fast. Alongside it, Eno, our general-purpose robot, is moving from engineering builds toward production, and the first units will be assembled, tested and maintained here. You are the first lead technician for both: you keep the collection floor running today, and you are the person who gets to build Enos from scratch as production starts, and to grow the technician team that follows you.
When a robot arm stops, a hand loses tracking, a camera drifts out of calibration or a cable fails, data stops, and you are the one who gets it running again, finds out why, and makes sure it does not come back. When a new station or a new site has to be set up, you build it. When a task needs a fixture that does not exist, you design it. You work with the operators every day, with the engineers who design the hardware, and with the reliability team when they test on the floor.
Keep the collection floor running: diagnose, repair and maintain robot arms, hands, wheeled bases, station computers and cameras, and the wearable collection kits (head-mounted cameras, gloves, wireless hubs); swap, repair, verify, return to service.
Assemble and bring up Eno units as production starts: build from assembly drawings and work instructions, run the end-of-line checks, and maintain the fleet once it is running.
Set up stations and sites: build new stations from the station BOM, reconfigure them when the layout changes, and set up stations for deployments at partner and customer sites so they run the same way they run here.
Design and build mini fixtures, jigs, brackets and cable harnesses: crimped and soldered connectors to a workmanship standard, built to drawing or to your own sketch when the drawing does not exist yet.
Run calibration and readiness: arm, hand and camera calibration on cadence and after any repair; daily station checks; the tests that say a station is ready for collection or evaluation.
Find the root cause: every failure gets a tag from our failure list and a short write-up; repeat failures get a real investigation with the engineers and a fix that sticks.
Keep the records: unit, serial, what failed, what was done, parts used, in our fleet and maintenance system; spares and consumables stocked and reordered before they run out.
Support the engineers: real-robot testing with the reliability team, retrofits and hardware changes on the floor, feedback on what breaks and why.
Write it down: repair procedures, calibration SOPs, station build instructions with photos, so the next technician does it the same way.
Grow the team: train the technicians who join after you and set the standard for how work is done.
Five or more years as an electromechanical, robotics or test technician in a lab, pilot line or production environment. Robots, automation, vehicles, medical devices or lab instruments all count.
You read assembly drawings, schematics and wiring diagrams and build from them.
You build cables and fixtures: crimping, soldering, connector assembly, strain relief, basic fabrication, and you can design a simple fixture when none exists.
You troubleshoot with instruments: multimeter, oscilloscope basics, continuity and power checks, and you are comfortable at a Linux command line to read logs, restart services and run scripts.
You have run calibration or alignment procedures on robots, cameras or sensors and know why cadence and documentation matter.
You think in root cause, not symptoms, and you write it down.
You are organized about parts, tools and records, and you notice when something is off.
You have trained or led other technicians, or you are ready to.
Bonus: robot arms or dexterous hands; 3D printing and basic machining; electrical safety training; time on a manufacturing or NPI line.
Production and lab environments; prolonged standing; lifting and moving station equipment; occasional travel to partner and customer sites.
You will be one of the first technicians to build our robots, and the floor only produces when the hardware runs. What you learn about how it fails shapes the next version.
We are always looking for people who are driven to work on hard problems and want to shape what the world will look like in ten years. A few of the values that guide us:
Limitless ambition. We hold ourselves to the highest bar in everything we build.
Relentless urgency. Iterate faster, learn faster. Time is the denominator.
Type-2 Fun. Take the work seriously, not yourself. Embrace the grind and find joy in the struggle.
Radical transparency. Candid communication builds trust, and direct feedback drives growth.
Multiculturalism. We come from across the world. Be curious. Meet people on their terms.
Team over ego. Help others, share context, and win as a team.
A joyful journey. We take pride in the craft, enjoy the ride, and make space to laugh. Happiness is part of how we get there.
We are a full-stack general-purpose robot company, born across San Francisco and Paris and bound by a single mission: to make general-purpose robots a reality, unlocking infinite physical labor for the world and freeing people for creativity, exploration, and the pursuits they love for their own sake.
We are French and American by birth and global by nature. We are backed by partners who share our vision, including Eclipse, Khosla Ventures, Bpifrance, and HSG, alongside Eric Schmidt, Xavier Niel, Daniela Rus, and Vladlen Koltun. Our people work across the Bay Area and Europe, building the engine that will teach robots to do real work in the real world. We recently unveiled Eno, our wheeled, dexterous robot powered by our GENE foundation model, and we are starting to put it to work with real customers.
Own the technical path from customer use case to working deployment, turning real-world workflows into solutions our robots can reliably execute.
Work directly with customers to understand their operations, assess automation feasibility, and define requirements, success metrics, and deployment scope.
Design and implement deployment solutions across product management, robot hardware, model capabilities, tooling, and customer systems, making practical tradeoffs between performance, reliability, and integration complexity.
Get hands-on at customer sites: configure systems, run evaluations, debug failures, and drive iterations until the solution meets agreed acceptance criteria.
Work closely with research, robotics engineering, and platform teams to turn field failures into actionable improvements, distinguishing what requires a better model, a software fix, or a change to the physical setup.
Build reusable tools, integration patterns, and deployment playbooks that make each new site faster and easier to bring online.
Track record of delivering technical solutions in real-world environments, with ownership that extends through integration, troubleshooting, and customer acceptance.
Strong hands-on engineering skills, including Python, comfort working in Linux, and experience integrating hardware and software through APIs, interfaces, or control systems.
Enough robotics depth to investigate a failed task across system.
Ability to translate an ambiguous customer request into a concrete technical plan, explain tradeoffs clearly, and set expectations grounded in what the system can deliver.
Strong bias toward the field over the whiteboard, with the persistence to work through the gap between a successful demo and reliable daily operation.
Clear written communication skills to document configurations, failure modes, and operating procedures that others can use without you in the room.
Comfort operating with incomplete information and traveling to customer sites to support deployments.
Bonus: Experience with robot manipulation, ROS/ROS 2, industrial automation, or integrating systems into manufacturing and logistics workflows.
We are a full-stack general-purpose robot company, born across San Francisco and Paris and bound by a single mission: to make general-purpose robots a reality, unlocking infinite physical labor for the world and freeing people for creativity, exploration, and the pursuits they love for their own sake.
We are French and American by birth and global by nature. We are backed by partners who share our vision, including Eclipse, Khosla Ventures, Bpifrance, and HSG, alongside Eric Schmidt, Xavier Niel, Daniela Rus, and Vladlen Koltun. Our people work across the Bay Area and Europe, building the engine that will teach robots to do real work in the real world. We recently unveiled Eno, our wheeled, dexterous robot powered by our GENE foundation model, and we are starting to put it to work with real customers.
Turn customer workflows into clearly scoped deployments, using site visits and operator interviews to define requirements, success metrics, and acceptance criteria.
Build deployment financial models and assess customer value, utilization, implementation costs, and support needs to inform scope and rollout decisions.
Partner with Deployment Solution Engineers, research, engineering, and safety specialists to align technical capabilities with customer requirements and site readiness.
Drive cross-functional execution from scoping through customer acceptance, keeping milestones, dependencies, scope changes, and decisions clear across distributed teams.
Coordinate acceptance testing, training, and handover, and turn field feedback and competitive insights into reusable deployment practices.
A track record of taking complex technical initiatives from ambiguous customer needs through launch and adoption.
Strong analytical judgment, with experience building financial models and evaluating costs, resource constraints, and implementation tradeoffs.
Enough technical fluency to work with robotics, software, and AI teams, question assumptions, and translate field issues into clear requirements.
A strong bias toward the field, with the ability to learn directly from operators and communicate requirements, decisions, and tradeoffs clearly in writing.
Comfort driving alignment and execution with incomplete information across customers, engineers, and external partners.
Bonus: Experience in robotics, autonomy, industrial technology, or enterprise systems where success depends on reliable real-world performance.
We are a full-stack general-purpose robot company, born across San Francisco and Paris and bound by a single mission: to make general-purpose robots a reality, unlocking infinite physical labor for the world and freeing people for creativity, exploration, and the pursuits they love for their own sake.
We are French and American by birth and global by nature. We are backed by partners who share our vision, including Eclipse, Khosla Ventures, Bpifrance, and HSG, alongside Eric Schmidt, Xavier Niel, Daniela Rus, and Vladlen Koltun. Our people work across the Bay Area and Europe, building the engine that will teach robots to do real work in the real world. We recently unveiled Eno, our wheeled, dexterous robot powered by our GENE foundation model, and we are starting to put it to work with real customers.
You own the supply chain for a general-purpose robot going from prototype to production: every part in the bill of materials, from actuators, gears and motors to structures, PCBAs, harnesses, batteries and sensors, and every supplier and manufacturing partner behind them. You build it from the ground up, in the United States and abroad, and you build the team that runs it. The scale goes from tens of units to hundreds to thousands, and the supply base, partners and processes that got us to today will not be the ones that get us there. Building the next version is the job.
This is a role for someone who has done it before, in the trenches, and wants to do it again with a product that matters more than the last one. You will spend your weeks with suppliers, on the floor with our manufacturing partners, in design reviews with the engineers who are still changing the parts, and with the leadership team on the decisions that shape the company. Strategy and execution are the same job here.
Few people get to decide how a new category of machine gets built. Eno is dexterous, wheeled and already working with real customers, built today with a supply base and partners that got us here. The next version of that supply chain, the one that takes us to volume, is yours to shape: what we make and what we buy, where in the world it comes from, which partners we grow with, and what the robot costs at scale. Those choices will be visible in every unit we ship for years.
Strategy. Set the supply chain strategy from pilot to volume: make versus buy, regional footprint, domestic content for regulated markets, single versus dual source by sub-system, and the order in which each piece moves as volume grows.
Partners. Own contract manufacturer and partner relationships at the executive level: selection, agreements, capacity commitments, cost and quality performance, and the transfer of sub-assemblies between regions and partners when the plan calls for it.
Cost. Own the cost roadmap: should-cost and total-cost models, the path from pilot cost to target cost, tooling and capital decisions, long-term agreements with pricing that steps down with volume; partner with Finance on working capital and with Legal on supplier agreements.
Supply base. Build the supplier base for the hard sub-systems: actuators, gears and motors, machined and cast structures, PCBAs, harnesses, shells, batteries and sensors; two qualified sources for every critical part, in the regions the product and our customers require.
Risk. Run capacity and risk: capacity ahead of demand, lead-time and allocation strategy for constrained parts, tariff and trade exposure, geopolitical and single-point-of-failure risk, and a plan for each.
Team. Build and lead the organization: sourcing, supplier development, planning, supplier quality and logistics. Hire the leaders under you, hold a high bar, set the operating cadence and the weekly numbers (clear-to-build, on-time delivery, purchase price variance, inventory, supplier quality) and the readout to the leadership team and the board.
Systems. Stand up what a scaled organization runs on, with Manufacturing Engineering and Platform: part numbering, BOM and change control, MRP and ERP, supplier portal, traceability. Lightweight first, then more.
Design. Be in the design conversation from the start: bring supplier capability and cost into the product early, and push the design toward what can be built at volume at the right price.
Built it before. You have led the supply chain organization for a complex electromechanical product from first builds through production at scale, and either built the supply base from scratch or moved it to a new geography. Robotics, electric vehicles, aerospace, consumer electronics or medical devices.
Leader of leaders. You have built a supply chain organization, hired its managers, and run it on numbers; you can still do the work yourself when the team is three people.
Commercial weight. You have negotiated with contract manufacturers and tier-one suppliers at the executive level and closed long-term agreements that held up under pressure.
Global. You have sourced across regions, including Asia and North America, and moved production between them.
Cost record. You can walk through a BOM cost-down from pilot to volume part by part, and you know where the money hides.
Technical. You read drawings and GD&T, you understand actuators, gearboxes, motors and PCBAs well enough to challenge an engineer constructively, and you are fluent in BOMs, change control, should-cost and MRP.
Pace. You default to urgency, operate without a playbook, and are comfortable before the systems exist.
Experience. Typically 10 or more years in supply chain, sourcing or manufacturing operations for complex hardware, with time in a mature organization so you know what good looks like at scale, and time in a company that was building it. Engineering degree preferred.
On-site in the Bay Area. Travel to suppliers and partner sites, including international, up to 40%.
We are always looking for people who are driven to work on hard problems and want to shape what the world will look like in ten years. A few of the values that guide us:
Limitless ambition. We hold ourselves to the highest bar in everything we build.
Relentless urgency. Iterate faster, learn faster. Time is the denominator.
Type-2 Fun. Take the work seriously, not yourself. Embrace the grind and find joy in the struggle.
Radical transparency. Candid communication builds trust, and direct feedback drives growth.
Multiculturalism. We come from across the world. Be curious. Meet people on their terms.
Team over ego. Help others, share context, and win as a team.
A joyful journey. We take pride in the craft, enjoy the ride, and make space to laugh. Happiness is part of how we get there.
","descriptionPlain":"ABOUT GENESIS\n\nWe are a full-stack general-purpose robot company, born across San Francisco and Paris and bound by a single mission: to make general-purpose robots a reality, unlocking infinite physical labor for the world and freeing people for creativity, exploration, and the pursuits they love for their own sake.\n\nWe are French and American by birth and global by nature. We are backed by partners who share our vision, including Eclipse, Khosla Ventures, Bpifrance, and HSG, alongside Eric Schmidt, Xavier Niel, Daniela Rus, and Vladlen Koltun. Our people work across the Bay Area and Europe, building the engine that will teach robots to do real work in the real world. We recently unveiled Eno, our wheeled, dexterous robot powered by our GENE foundation model, and we are starting to put it to work with real customers.\n\n\n\n\nTHE ROLE\n\nYou own the supply chain for a general-purpose robot going from prototype to production: every part in the bill of materials, from actuators, gears and motors to structures, PCBAs, harnesses, batteries and sensors, and every supplier and manufacturing partner behind them. You build it from the ground up, in the United States and abroad, and you build the team that runs it. The scale goes from tens of units to hundreds to thousands, and the supply base, partners and processes that got us to today will not be the ones that get us there. Building the next version is the job.\n\nThis is a role for someone who has done it before, in the trenches, and wants to do it again with a product that matters more than the last one. You will spend your weeks with suppliers, on the floor with our manufacturing partners, in design reviews with the engineers who are still changing the parts, and with the leadership team on the decisions that shape the company. Strategy and execution are the same job here.\n\n\n\n\nWHY THIS ROLE\n\nFew people get to decide how a new category of machine gets built. Eno is dexterous, wheeled and already working with real customers, built today with a supply base and partners that got us here. The next version of that supply chain, the one that takes us to volume, is yours to shape: what we make and what we buy, where in the world it comes from, which partners we grow with, and what the robot costs at scale. Those choices will be visible in every unit we ship for years.\n\n\n\n\nWHAT YOU'LL DO\n\nStrategy. Set the supply chain strategy from pilot to volume: make versus buy, regional footprint, domestic content for regulated markets, single versus dual source by sub-system, and the order in which each piece moves as volume grows.\n\nPartners. Own contract manufacturer and partner relationships at the executive level: selection, agreements, capacity commitments, cost and quality performance, and the transfer of sub-assemblies between regions and partners when the plan calls for it.\n\nCost. Own the cost roadmap: should-cost and total-cost models, the path from pilot cost to target cost, tooling and capital decisions, long-term agreements with pricing that steps down with volume; partner with Finance on working capital and with Legal on supplier agreements.\n\nSupply base. Build the supplier base for the hard sub-systems: actuators, gears and motors, machined and cast structures, PCBAs, harnesses, shells, batteries and sensors; two qualified sources for every critical part, in the regions the product and our customers require.\n\nRisk. Run capacity and risk: capacity ahead of demand, lead-time and allocation strategy for constrained parts, tariff and trade exposure, geopolitical and single-point-of-failure risk, and a plan for each.\n\nTeam. Build and lead the organization: sourcing, supplier development, planning, supplier quality and logistics. Hire the leaders under you, hold a high bar, set the operating cadence and the weekly numbers (clear-to-build, on-time delivery, purchase price variance, inventory, supplier quality) and the readout to the leadership team and the board.\n\nSystems. Stand up what a scaled organization runs on, with Manufacturing Engineering and Platform: part numbering, BOM and change control, MRP and ERP, supplier portal, traceability. Lightweight first, then more.\n\nDesign. Be in the design conversation from the start: bring supplier capability and cost into the product early, and push the design toward what can be built at volume at the right price.\n\n\n\n\nWHAT WE ARE LOOKING FOR\n\nBuilt it before. You have led the supply chain organization for a complex electromechanical product from first builds through production at scale, and either built the supply base from scratch or moved it to a new geography. Robotics, electric vehicles, aerospace, consumer electronics or medical devices.\n\nLeader of leaders. You have built a supply chain organization, hired its managers, and run it on numbers; you can still do the work yourself when the team is three people.\n\nCommercial weight. You have negotiated with contract manufacturers and tier-one suppliers at the executive level and closed long-term agreements that held up under pressure.\n\nGlobal. You have sourced across regions, including Asia and North America, and moved production between them.\n\nCost record. You can walk through a BOM cost-down from pilot to volume part by part, and you know where the money hides.\n\nTechnical. You read drawings and GD&T, you understand actuators, gearboxes, motors and PCBAs well enough to challenge an engineer constructively, and you are fluent in BOMs, change control, should-cost and MRP.\n\nPace. You default to urgency, operate without a playbook, and are comfortable before the systems exist.\n\nExperience. Typically 10 or more years in supply chain, sourcing or manufacturing operations for complex hardware, with time in a mature organization so you know what good looks like at scale, and time in a company that was building it. Engineering degree preferred.\n\n\n\n\nWHAT ELSE YOU NEED TO KNOW\n\nOn-site in the Bay Area. Travel to suppliers and partner sites, including international, up to 40%.\n\n\n\n\nHOW WE WORK\n\nWe are always looking for people who are driven to work on hard problems and want to shape what the world will look like in ten years. A few of the values that guide us:\n\nLimitless ambition. We hold ourselves to the highest bar in everything we build.\n\nRelentless urgency. Iterate faster, learn faster. Time is the denominator.\n\nType-2 Fun. Take the work seriously, not yourself. Embrace the grind and find joy in the struggle.\n\nRadical transparency. Candid communication builds trust, and direct feedback drives growth.\n\nMulticulturalism. We come from across the world. Be curious. Meet people on their terms.\n\nTeam over ego. Help others, share context, and win as a team.\n\nA joyful journey. We take pride in the craft, enjoy the ride, and make space to laugh. Happiness is part of how we get there."},{"id":"7a6b9277-5d73-494b-ad72-dd5fe2ccf763","title":"Social Media Manager","department":"Product & Marketing","team":"Product & Marketing","employmentType":"FullTime","location":"Bay Area","secondaryLocations":[],"publishedAt":"2026-10-24T03:10:05.025+00:00","isListed":true,"isRemote":false,"workplaceType":"OnSite","address":{"postalAddress":{"addressRegion":"California","addressCountry":"United States","addressLocality":"San Carlos"}},"jobUrl":"https://jobs.ashbyhq.com/genesis/7a6b9277-5d73-494b-ad72-dd5fe2ccf763","applyUrl":"https://jobs.ashbyhq.com/genesis/7a6b9277-5d73-494b-ad72-dd5fe2ccf763/application","descriptionHtml":"We are a full-stack general-purpose robot company, born across San Francisco and Paris and bound by a single mission: to make general-purpose robots a reality, unlocking infinite physical labor for the world and freeing people for creativity, exploration, and the pursuits they love for their own sake.
We are French and American by birth and global by nature. We are backed by partners who share our vision, including Eclipse, Khosla Ventures, Bpifrance, and HSG, alongside Eric Schmidt, Xavier Niel, Daniela Rus, and Vladlen Koltun. Our people work across the Bay Area and Europe, building the engine that will teach robots to do real work in the real world. We recently unveiled Eno, our wheeled, dexterous robot powered by our GENE foundation model, and we are starting to put it to work with real customers.
You'll own Genesis's brand across social platforms, starting with X, and build awareness, understanding, and engagement among researchers, labs, scientists, and prospective hires. Working with engineering and founders, you'll capture research, hardware, and model-training work and show the evidence behind Genesis's leadership in the field. This hire matters now because deployment brings more real work to share and more scrutiny of every claim, requiring accurate publishing and support for founder-led content.
Run Genesis's social channels, starting with X and extending to LinkedIn, YouTube, and other platforms, owning the publishing calendar and adapting material to each audience.
Capture and edit short videos from the lab and field, or direct the shoot and edit, planning footage with engineers so viewers can see the task and understand the result.
Turn research and model-training work into technical threads built around a specific finding, evaluation choice, hardware constraint, or lesson from a failed attempt.
Spend time in engineering reviews and with founders each week to capture work at the source and develop founder-led posts that preserve individual voices and technical judgment.
Engage researchers and the Genesis World open-source community through substantive replies and questions, crediting contributors and routing technical questions to the right people.
Coordinate launch-day publishing with product marketing, support employee voices, and prepare assets that make accurate sharing easy.
Review robot footage and captions for platform, operating conditions, autonomy, playback speed, and edits that affect interpretation, confirming customer approval and data-rights review for customer material.
Measure reach and meaningful engagement among researchers, engineers, and target candidates, using audience quality and results to improve the next week's work.
Social media or editorial experience, with direct responsibility for a technical organization's channels.
Technical comprehension sufficient to find a useful story in a research paper or hardware review while preserving the caveats that matter.
Short-form video production skills, including hands-on shooting and editing or experience directing others with specific briefs and useful feedback.
Strong writing skills across concise posts and substantive technical threads, with judgment about when a diagram or an unbroken clip explains more than copy.
Sound editorial judgment about what is ready to publish, what needs technical review, and what should stay private.
Track record of building relationships with researchers and open-source contributors through useful exchanges and consistent credit.
Clear understanding of channel analytics and audience quality, supported by examples of changes made in response to evidence.
Bonus: Experience covering robotics or embodied AI, with familiarity with open-source research communities and contributor norms.
Bonus: A record of supporting founder-led content while preserving individual voices.
Bonus: Experience handling public questions about automation and jobs with direct answers and respect for industrial workers.
We are a full-stack general-purpose robot company, born across San Francisco and Paris and bound by a single mission: to make general-purpose robots a reality, unlocking infinite physical labor for the world and freeing people for creativity, exploration, and the pursuits they love for their own sake.
We are French and American by birth and global by nature. We are backed by partners who share our vision, including Eclipse, Khosla Ventures, Bpifrance, and HSG, alongside Eric Schmidt, Xavier Niel, Daniela Rus, and Vladlen Koltun. Our people work across the Bay Area and Europe, building the engine that will teach robots to do real work in the real world. We recently unveiled Eno, our wheeled, dexterous robot powered by our GENE foundation model, and we are starting to put it to work with real customers.
There's more real work happening at Genesis each week than anyone outside the building gets to see. With a launch ahead and researchers and engineers watching closely, we want a steady, credible record of it: the lab, the robots, the people, and the days between demos. The In-House Creator is our production engine, spotting the story, shooting it, and finishing it. You'll work with the Social Media Manager, who owns strategy and the calendar, and with founders and engineers, to publish often without giving up the premium feel.
Capture everyday life at Genesis, from late test runs to whiteboard sessions to lunch, and turn it into a steady pipeline of material for the brand accounts.
Film robots and lab work safely and closely enough that the footage shows real capability, favoring unbroken takes on real tasks over staged moments.
Shoot, produce, and edit video end to end, handling camera, lighting, audio, pacing, music, and the final cut, with each version built natively for its platform and X first.
Develop recurring formats, such as lab updates, engineer explainers, and first-attempt robot reels, that the team can count on and the audience comes back for.
Put founders, researchers, and engineers on camera, helping people who'd rather be debugging come across clearly and like themselves.
Cover launches, events, and lab visits, and turn customer deployments into video stories as robots go into production.
Build a searchable footage library and a capture system with shot lists and filming guides so engineers and field teams bring back usable footage when you're not there.
Work with the Social Media Manager on what ships, when, and in what form, supplying both the raw and finished material behind the calendar.
Review every robot clip for accuracy by labeling autonomy level and playback speed, flagging any edit that changes how a result reads, and holding customer and unreleased work until it's cleared.
Keep the bar high while publishing often, deciding what stays inside, what gets cut, and what earns a place on the brand accounts.
Professional experience (5+ years) shooting, producing, and editing video, with a portfolio that shows both what you captured and how you cut it.
Hands-on craft with camera, lighting, and audio, including fast setups in tight lab spaces and upkeep of the kit you shoot with.
Fluency in Premiere Pro, DaVinci Resolve, Final Cut, CapCut, or similar, plus a working sense of hooks in the first seconds, pacing, and sound for short-form video.
Speed paired with taste, meaning the ability to turn a moment around the same day without it looking rushed.
Comfort on camera and a presence that fits Genesis: curious, credible, and understated.
Genuine curiosity about robotics and the patience to work around engineers, test schedules, and hardware that doesn't always cooperate.
Safety awareness around moving robots and active test areas, and the habit of asking before stepping into a work cell.
Sound judgment about what's ready to publish, including when a clip overstates what a robot can do or reveals something that should stay inside.
Bonus: An existing following whose tone and audience align with Genesis.
Bonus: Motion graphics, 3D, or AI-assisted creative tools that help you move faster without lowering the bar.
Bonus: Prior work in robotics, hardware, or another field where a lab demo and a field deployment are very different things to film.