Push the frontier of how machines read and reason about law, then ship it. You will turn open problems (long contracts, cross-document reasoning, faithful citations, multilingual precedent) into results that reach lawyers inside Word within a quarter, not a paper cycle.

What you will do

• Research long-context and retrieval-augmented reasoning over contract sets of hundreds of documents, such as a full data room.
• Build benchmarks with practising lawyers that measure what matters: missed risks, false alarms, faithfulness of cited reasoning.
• Study failure modes of LLM judges and design evaluation that a general counsel would trust.
• Prototype agents for Trial & Arbitration Simulation that argue both sides of a case and stay grounded in the record.
• Publish or open-source where it helps the field, with the customer's data never leaving their control.

What you bring

• PhD or equivalent research record in NLP, information retrieval or machine learning, with first-author work at venues such as ACL, EMNLP, NeurIPS or SIGIR.
• Fluent experimental practice in Python and PyTorch or JAX; you write code that others can rerun.
• Evidence that you have taken research into a product or a widely used tool.
• Curiosity about law. A legal background is welcome but not required.

Nice to have

• Work on evaluation methodology, LLM-as-judge, calibration or uncertainty estimation.
• Multilingual or cross-lingual retrieval, especially across European languages and Chinese.

How we hire

A 10-minute AI interview in English (you are told up front that you are speaking with an AI), a conversation with the team you would join, and a short paid practical task that mirrors real work. We aim to decide within three weeks.