Research Engineer / Scientist, Reinforcement Learning

Role overview Advance post-training methods for language models used in legal reasoning and agentic workflows. You’ll own research from experiment design through scaled training and evaluation, turning improvements to open-weight models into capabilities used in production.

Responsibilities - Set and pursue research directions for long-horizon legal reasoning and agentic tasks. - Design and run supervised fine-tuning and reinforcement learning experiments on open-weight models. - Develop and assess rewards and verifiers for tasks with partially checkable outcomes, using rubrics, learned preference models, and expert feedback. - Run and debug large-scale training, investigating instability, divergence, and reward hacking. - Build rigorous evaluations to measure progress and inform release decisions. - Implement, test, and debug code across the research stack.

Requirements - Direct experience post-training language models with supervised fine-tuning and reinforcement learning, including preference optimization. - Ownership of training runs and their results. - Strong machine learning fundamentals and experience making distributed training reliable. - Strong Python engineering skills, with experience delivering production-quality code and owning systems end to end. - Research experience in language-model reinforcement learning or reward modeling, and ability to move from experiments to shipped improvements.

Benefits and work setup - The role is open across multiple levels. - Listed compensation is $230,000–$450,000 USD.