Research Engineer / Scientist, Reinforcement Learning
HarveyVisit Harvey website
Harvey is an AI platform for legal and professional-services organizations.
Series F+Recently funded
Distributed
About Harvey
Harvey provides agents and collaborative tools for legal work, including legal research, document analysis, secure workspaces, contract intelligence, and organization-level analytics.
Skills
Candidate Availability
Required and preferred rules are kept separate and reflect the wording in the original posting.
About the Role
You will own how Harvey adapts and improves open-weight models for legal work. You'll design and run SFT and RL post-training experiments, develop rewards and verifiers for semi-verifiable tasks, debug large-scale training, and carry ideas through to models used by customers.
Requirements
- Hands-on experience post-training LLMs with SFT and RL including preference optimization
- Deep ML fundamentals and experience with distributed training runs
- Strong Python engineering skills
- Research experience in RL for language models or reward modeling
Responsibilities
- Own a research agenda to improve long-horizon legal reasoning and agentic workflows
- Design and run post-training experiments across SFT and RL
- Develop reward functions and verifiers for semi-verifiable tasks
- Run and debug training at scale, addressing divergence, reward hacking and instability
- Build rigorous evals to track improvements and decide what ships
- Design, implement, test and debug code across the research stack
Benefits
- Equity