Want to know if this job is worth applying to?
San Francisco, California, United States
Onsite
Introducing Moonlake, AI for creating world simulations.
Moonlake is building the frontier of interactive world models: systems that generate, simulate, and reason over 3D environments for robotics, embodied AI, and interactive applications.
We develop the infrastructure that enables intelligent systems to learn, evaluate, and interact within realistic virtual environments before operating in the physical world.
Our work sits at the intersection of:
Robotics
Embodied AI
Interactive 3D Worlds
World Models
Simulation Infrastructure
Physical AI
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, physics, and intelligent behavior.
Our team has raised $50M 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 and simulation to digital twins and interactive environments.
We are looking for exceptional engineers to help build the simulation systems that will power the next generation of robotics and embodied intelligence.
We are hiring a Member of Technical Staff to lead reinforcement learning infrastructure and agentic model post-training.
You will work closely with Qi and the research team to improve large vision-language and code-generating agents through fine-tuning, reinforcement learning, trajectory data, and scalable evaluation.
Moonlake already has deep expertise in 3D and world-building. This role adds the model-training experience needed to systematically improve agent performance and prepare the company for larger-scale RL across both digital and physical environments.
We are looking for a full-stack researcher and engineer who understands the complete training system and can make strong judgments about when training is necessary, which methods are likely to work, and what not to pursue.
Build RL and post-training pipelines for multimodal, vision-language, and code-generating agents
Develop infrastructure for supervised fine-tuning, preference optimization, reward modeling, and reinforcement learning
Create systems for collecting, filtering, replaying, and learning from agent trajectories
Design rewards, verifiers, and evaluations for long-horizon agent tasks
Improve agents’ ability to plan, write and execute code, use tools, recover from errors, and complete complex workflows
Scale distributed training and high-throughput rollout generation across multi-GPU environments
Improve training reliability, reproducibility, observability, and cost efficiency
Help define Moonlake’s long-term strategy for agent, robotics, and embodied-model training
Real-world experience training large language, vision-language, multimodal, or code models
Strong experience in reinforcement learning, post-training, or large-scale fine-tuning
Experience building distributed training or high-throughput inference systems
Familiarity with supervised fine-tuning, preference optimization, reward modeling, and agentic RL
Experience with code-generation agents, long-horizon evaluation, or tool-using systems
Strong Python skills and experience with PyTorch, JAX, or similar frameworks
Ability to work across data, models, environments, rewards, evaluation, and infrastructure
Strong research judgment and a bias toward building reliable systems
Experience at a frontier AI lab or organization operating large-scale training systems
Experience with code-model post-training or autonomous coding agents
Experience with multimodal models, robotics, simulation, or embodied AI
Experience designing verifiable rewards or outcome-based training systems
Experience scaling RL workloads across large GPU clusters
Within your first year, you will have:
Built Moonlake’s core post-training and RL infrastructure
Created scalable systems for learning from agent trajectories
Delivered measurable improvements in agent quality and task completion
Helped the team determine which problems require training and which do not
Established a reliable foundation for larger-scale agent and embodied-model training
Moonlake’s agents must do more than generate content. They must understand complex requests, reason across vision and language, write and execute code, operate tools, build interactive worlds, and recover from mistakes.
This role will build the training systems that allow those agents to continuously improve.
We are committed to being an on-site, in-person team currently based in San Francisco.
