ML Researcher, Reinforcement Learning & Agents

Endo Health, Inc.; San Francisco (CA); On-site; USD 180,000 - 240,000; Full time

Job summary

Endo Health, Inc. is seeking a researcher to build the RL environment for self-improving medicine in a clinical setting. You will design branching care journeys, partial observations, and delayed outcomes to enable robust learning loops that improve patient care over time.

You’ll work with engineers and clinicians to define the problem, develop learning strategies, and validate improvements with unseen cases. San Francisco relocation support is offered.

Qualifications

• You've built and debugged training experiments yourself.
• You have depth in RL, post-training, or sequential decision-making.
• You can explain why a result is real, where it breaks, and what to test next.
• Shape the research agenda. Own the connection between the environment, objectives, training, and evaluation.

Responsibilities

• Build the environment: branching care journeys, tools, partial information, and delayed outcomes.
• Build the learning loop: demonstrations, reward design, post-training, and reproducible experiments.
• Prove improvement: beat baselines on unseen cases and catch reward hacking, leakage, and simulator shortcuts.

Skills

Reinforcement Learning

Job description

"Build the RL environment for self-improving medicine."

Endo is building self-driving healthcare, starting with GLP-1s. We believe patients should be able to delegate the work of getting treated to an AI that knows them over time. That means getting care delivered, following what happens, and using each experience to make the next one better.

The Role

Medicine unfolds over time. A decision today can produce feedback days or months later, and the outcome depends on much more than the recommendation. Did the patient receive treatment? Follow it? Tolerate it? What changed along the way? We think this is the experience AI needs to learn from.

Human demonstrations give us a starting point. Your job is to build the RL environment and learning loop that come next: what the agent can observe, which actions it can take, how consequences unfold, and what improvement actually means. You’ll work on partial observations, delayed rewards, and the gap between a convincing simulation and a policy that helps real patients.

You’ll help set the research agenda from the beginning, alongside the engineers building the product and the clinicians delivering care. The ambition is self-improving medicine. The work starts with rigorous environments, useful data, and experiments that tell us whether we’re getting closer.

You will:

• Build the environment: branching care journeys, tools, partial information, and delayed outcomes.


• Build the learning loop: demonstrations, reward design, post-training, and reproducible experiments.


• Prove improvement: beat baselines on unseen cases and catch reward hacking, leakage, and simulator shortcuts.

You've done this before if:

• You’ve built and debugged training experiments yourself.


• You have depth in RL, post-training, or sequential decision-making.


• You can explain why a result is real, where it breaks, and what to test next.


• Shape the research agenda. Own the connection between the environment, objectives, training, and evaluation.


• Help create the experience models learn from. Work with the people building and delivering care to capture decisions, execution, and outcomes over time.


• A founding team built for this problem. Founded by a medical doctor turned engineer and backed by leading Silicon Valley investors, including a16z speedrun, General Catalyst, and the founders of 23andMe, MyFitnessPal, and Spring Health.


• US relocation support. Join us in San Francisco.


• An experiment, paper, or repo you built. Include the baseline, result, and one finding that changed your mind.