Member of Technical Staff, Research Engineer (Reinforcement Learning)

About Epsilab Epsilab (YC F26) is an applied AI research lab building training simulations for real-world AI agents. We create high-fidelity reinforcement learning environments that recreate enterprise software, tools, and complex professional workflows, helping agents learn to complete long-horizon tasks. We’re a fast-growing startup working with frontier AI labs. Our team is extremely flat, fast-moving, and creative: you’ll work directly with the founders, contribute ideas across the company, and take ownership from first experiment to shipped product. We value technical curiosity, clear thinking, and people who turn ambitious ideas into working systems. About the role We’re hiring a Research Engineer focused on reinforcement learning to build environments, evaluation systems, and experimental infrastructure for training capable AI agents. You’ll investigate what makes agents succeed or fail on complex workflows, then turn those findings into better tasks, reward signals, and reliable research tools. You’ll work at the intersection of research and engineering, moving quickly from a hypothesis to a reproducible experiment and a system that frontier lab partners can use. Projects will require intense effort, attention to detail, and research taste. We are looking for people who will do whatever it takes to solve immediate bottlenecks, while also thinking in systems and building compounding loops. You will be at the frontier of AI development every day solving agentic knowledge work and the broad diffusion of AI. What you’ll do

• Collaborate with frontier lab researchers and domain experts to build stateful RL environments with realistic tools, application behavior, resets, and verifiable outcomes.
• Develop reward functions, graders, and evaluation harnesses that measure task completion and catch shortcuts, reward hacking, and brittle behavior.
• Design and run experiments on agent performance, task difficulty, generalization, and long-horizon reliability.
• Analyze agent trajectories and environment state to distinguish model failures from task, tooling, or grading issues.
• Build pipelines for generating, validating, and replaying tasks and trajectories, with quality checks that combine programmatic validation and expert judgment.

What you’ll bring

• 3+ years of relevant experience in research engineering, machine learning, or software engineering.
• Strong Python skills and experience building and debugging systems using Linux, Docker, and APIs.
• A practical understanding of reinforcement learning, including rewards, policies, exploration, and evaluation.
• Experience implementing ML experiments or agent systems, and sound experimental judgment: you can choose useful baselines, interpret noisy results, and identify misleading metrics.
• The initiative to tackle open-ended problems, communicate findings clearly, and ship in a fast-moving startup.

Helpful experience LLM post-training, RL algorithms, PyTorch or JAX, agent benchmarks, reward modeling, or automated environment validation. Research publications and open-source contributions are welcome; a PhD is not required. Compensation

• Full-time, in person in San Francisco, California.
• US visa sponsorship available.
• Annual base salary: $125,000–$300,000, plus meaningful equity.