Quick overview
Job description
About the Role
Join a healthtech team developing clinical decision-support tools for use in healthcare settings. You will build reinforcement learning environments and evaluations that help model-training teams improve clinical reasoning, with work validated alongside clinical experts.
What You'll Do
Build reinforcement learning environments for complex clinical reasoning, including multi-step tasks, tool use, state transitions, and measurable outcomes.
Design and calibrate reward functions, grading rubrics, and LLM judges, and test them for reward hacking and misleading scores.
Own the workflow from task design and data creation through environment implementation, evaluation, and delivery to model-training teams.
Run post-training and agent evaluation experiments, then use observed failures to improve environments.
Collaborate with clinicians and technical leadership to validate model outputs and improve the work.
What We're Looking For
A PhD in computer science, machine learning, or a related field, or at least 3 years of full-time reinforcement learning research or post-training experience; candidates should also have 3 or more years of research or industry experience, or be a final-year PhD student or postdoctoral researcher.
Hands-on experience building reinforcement learning environments for post-training or agent evaluation, and owning reward functions, grading rubrics, or LLM-judge calibration.
Production-level Python, PyTorch, and Transformers ecosystem skills, plus experience with LLMs, reinforcement learning, post-training, evaluation, or reward modeling.
A bachelor's degree or higher in computer science, machine learning, mathematics, or a related technical field.
Clear technical writing, including the ability to explain failed experiments and uncertainty, and a careful approach to validating plausible model outputs.
Comfort working across disciplines in an ambiguous environment and attending in-person offsites several times per year.
Clinical AI experience is useful. Knowledge of information retrieval, constraint solving, SAT/SMT, or domain-specific languages is also relevant.
Must be authorized to work in the United States or Canada. Visa sponsorship is not available.
Compensation & Benefits
Base salary: $250,000 to $400,000 USD annually. Visa sponsorship is not available.
Location
Remote, with the primary location in Canada. Candidates must be authorized to work in Canada or the United States.
Haystack's AI agent can fill in and submit this application for you. How AI auto apply works