Research Engineer / Scientist, Reinforcement Learning
$230k - $340k • Remote • Remote • FullTime
Posted 10h ago
Job Location
Remote
Tech Stack
Remote Work Policy
Fully remote
Employment Type
FullTime
Categories
AI Research Engineer
About the job
Harvey is transforming legal and professional services by combining agentic AI, an enterprise-grade platform, and deep domain expertise. This is a unique opportunity to join a fast-scaling company at an inflection point, defining a new category in real-time. The team is driven, ambitious, and committed to excellence, valuing decisiveness, simplicity, and continuous improvement. As a Member of Technical Staff in Harvey Research, you will adapt and improve leading open-weight models for legal work, focusing on post-training efforts including SFT and RL, reward design, and verification for complex legal tasks. You will contribute to setting research direction and bringing ideas from experimentation to customer-facing models within a high-bar, hands-on team.
Responsibilities
- Own and pursue a research agenda to improve model capability on long-horizon legal reasoning and agentic workflows.
- Design and run post-training experiments across SFT and RL, building on leading open-weight models.
- Develop reward functions and verifiers for semi-verifiable tasks, including rubrics, learned preference models, and expert preference data, and study their failure modes.
- Run and debug training at scale, addressing issues like divergence, reward hacking, and instability.
- Build and utilize rigorous evaluations to track improvements and determine which models to ship.
- Design, implement, test, and debug code across the research stack.
Requirements
- Hands-on experience post-training LLMs with SFT and RL, including preference optimization, with direct ownership of training runs and results.
- Deep ML fundamentals and experience reliably running distributed training jobs.
- Strong engineering skills in Python, capable of writing production-quality code and owning end-to-end systems.
- Research experience in RL for language models or reward modeling, with the ability to rapidly translate experiments into shipped improvements.
- Publications are welcome but not required.