Retrieval & grounding
Finding the passage that answers a doctor's question, and keeping every answer tied to its source.
We study how medical AI finds the right evidence, answers from it, and is tested before it reaches doctors.
Finding the passage that answers a doctor's question, and keeping every answer tied to its source.
Choosing and adapting open models for reliable medical search and answers at a sensible cost.
Strict scoring, held-aside questions and measured run-to-run variation, so progress is real.
Tests built to break a feature before release, clinician review, and asking rather than guessing.