Member of Technical Staff, Research Engineer

About Crosby Crosby is built on a simple conviction: a great legal system is the watermark of a great society — and the best legal work comes from combining human expertise with the right technology, not replacing one with the other. Crosby is the first AI-native law firm, helping ambitious companies like Cursor, Ramp, and Cognition sign commercial contracts faster. Legal work is both art and science, and we're mapping the frontier between the two — codifying what can be systematized, amplifying human judgment where it matters most. That's why the right way to bring AI into law isn't to sell software and walk away. It's to own the outcome together: higher-quality work, delivered faster. We build proprietary tools and human-in-the-loop workflows that change what's possible for corporate legal teams. Crosby was founded by Ryan (Stanford Law, Cooley, startup GC) and John (Penn, first 15 engineers at Ramp). Help us transform one of society's most important industries. About the role As a Research Engineer, you'll bridge research and production. You'll turn validated methods, benchmarks, and evaluation infrastructure from our research team into dependable, production-grade features. You’ll be responsible for the entire lifecycle of an AI feature—from building the underlying systems to customizing the models that make them run. Your work will involve both robust system design a [... source excerpt omitted ...] sed with shipping reliable, high impact features. We also publish our work on legal AI at intelligence.crosby.ai — take a look at the kinds of problems you'd be working on. What you'll do Build core AI Systems: Design, build, and scale our primary AI features, including complex Retrieval Augmented Generation (RAG) pipelines and agentic workflows that make our models measurably better at contract review, playbook retrieval and docx editing. Build RL fine-tuning infrastructure: Stand up and harden RL training infrastructure (PPO, GRPO, and related methods) so researchers can train quickly and reliably at scale. Enable RL in a non-verifiable domain: Build reward pipelines, ge [... source excerpt omitted ...] ize Models for Impact: Selectively fine tune and adapt language models to improve their performance on highly specific legal tasks where off the shelf solutions fall short Ensure Production Quality: Develop evaluation and monitoring systems that catch regressions and reward hacking before they reach users. Apply AI Pragmatically: Stay on the frontier of AI research and identify the most effective techniques and tools to solve immediate product challenges. Collaborate to Ship: Work closely with product, backend, and our in house legal experts to deliver end to end solutions that make our users more effective. Who you are Experience: 3+ years of professional software engineering