Member of Technical Staff — RL & Agentic Model Training

Introducing Moonlake, AI for creating world simulations.
ABOUT MOONLAKE
Moonlake is building the frontier of interactive world models: systems that generate, simulate, and reason over 3D environments for embodied AI, robotics and gaming. We develop the simulation infrastructure to build worlds (e.g., assets, scenes, digital twins) at scale.
Our team sits at the intersection of:
- Embodied AI
- Robotics simulation
- Interactive 3D worlds
- World models
- Real-time generation
- AI infrastructure
Moonlake is building the next generation of AI infrastructure for interactive digital worlds. Our mission is to enable anyone to create, simulate, and interact with rich environments using natural language and multimodal inputs, turning simple ideas into worlds with structure, logic, and agents that can perceive and act.
Our team has raised $28M in seed funding from NVIDIA Ventures, Threshold Ventures, AIX ventures and notable angels including Naval Ravikant and Jeff Dean to build the foundational layer for the future of AI - powering everything from robotics training, simulations, and digital twins. Our goal is to make building and experimenting with these environments as accessible and scalable as publishing video on the internet.
We are looking for exceptional research engineers and applied researchers to help push the frontier of interactive AI.
THE ROLE
We're looking for an exceptional Member of Technical Staff focused on RL and Agentic Model Training to build and train models that learn through interaction.
You'll work across the full learning loop: training models, building RL environments in which they can learn, evaluating their behavior, and ultimately testing and improving these systems in the physical world.
You do not need to come from robotics (although it helps!). We're particularly interested in researchers and engineers who have trained agentic models, coding models, or other models using reinforcement learning and want to bring that expertise into embodied AI.
We want someone who can:
- Train and post-train agentic models using reinforcement learning
- Build environments and feedback loops that enable agents to learn complex tasks
- Develop training and data recipes across SFT, RL, and related post-training techniques
- Design reward functions, evaluations, and experiments that meaningfully improve model behavior
- Take ideas from research through implementation, large-scale experimentation, and real-world evaluation
WHAT YOU'LL DO
- Train and post-train agentic models using reinforcement learning, supervised fine-tuning, and related techniques.
- Build RL environments that allow models to interact, explore, receive feedback, and learn.
- Develop scalable training and data pipelines for agentic and embodied models.
- Design reward functions, model-based evaluations, and automated evaluation infrastructure.
- Develop new approaches for improving exploration, long-horizon reasoning, planning, and task completion.
- Run large-scale experiments and translate research ideas into measurable improvements in model performance.
- Work closely with world-model, simulation, infrastructure, and robotics teams.
- Evaluate trained models and policies in simulation and, increasingly, on real robotic systems.
- Use failures in simulation and the physical world to improve environments, training data, evaluation, and model behavior.
WHAT WE'RE LOOKING FOR
- Exceptional technical depth in machine learning, reinforcement learning, or agentic model training.
- Experience training or post-training large models using RL, SFT, or related techniques.
- Experience building environments in which agents interact and learn—not just training models on static datasets.
- Strong understanding of modern reinforcement learning, reward design, evaluation, and experimentation.
- Ability to build research infrastructure and production-quality training systems, not just develop algorithms.
- Strong software engineering skills in Python and modern ML frameworks.
- Track record of exceptional research, engineering, or model performance.
- Ability to operate in highly ambiguous research environments and take ideas from first principles to working systems.
WHY THIS ROLE MATTERS
The next generation of AI systems won't just generate outputs—they'll take actions, interact with environments, learn from feedback, and operate in the physical world.
That requires more than better models. It requires environments where agents can practice, evaluation systems that determine whether they're succeeding, and training loops that turn those interactions into better behavior.
Moonlake is building that infrastructure for Physical AI.
You'll help connect model training, RL environments, simulation, and real-world robotics, building systems that allow increasingly capable agents to learn complex behaviors before—and while—they operate in the physical world.
BONUS POINTS
- Experience in any of the following is valuable, but robotics experience is not required:
- Embodied AI or robot learning
- Vision-language-action models
- Sim-to-real or real-to-sim workflows
- Robotics simulation platforms such as Isaac Sim or MuJoCo
- World models and learned simulators
- Training or evaluating policies on physical robots
We are committed to being an on-site, in-person team currently based in San Francisco.

Employment type
FullTime

Work arrangement
No