Clinical Guideline Assistant (RAG Prototype)

Working in health tech taught me that the hard part of AI in care is trust, not fluency. This personal R&D prototype explores a narrow, safer pattern: an assistant that may only answer from a curated set of clinical guideline documents and must show the exact passages it relied on. Documents are chunked, embedded and stored in a vector index; each question retrieves the most relevant passages, and the model is instructed to answer only from them or say it does not know. Every response carries citations back to the source document and section. The prototype is built around evaluation: a fixed question set with expected source passages measures retrieval accuracy and answer faithfulness on every change. It is a decision-support experiment, not a diagnostic tool, and keeps a human in the loop by design.