Clinical Question Answering Agent with Evidence and Abstention

A clinical question answering agent that retrieves textbook and PubMed evidence (FAISS + BM25), answers a multiple-choice question with cited passages and a confidence (DSPy ChainOfThought with BootstrapFewShot demos), lets an independent critic check whether the evidence supports the answer, revises once, and ABSTAINS - escalating to a clinician approval queue - when the evidence does not support the answer or the confidence is below 60. The model qwen2.5:3b and the EmbeddingGemma embeddings are served by Ollama on a second laptop (RTX 2050 4 GB) over Wi-Fi; the code runs on a laptop with RTX 2050 4 GB (Ryzen 7, 8 GB RAM).

DISCLAIMER: Decision support only - not medical advice. Answers come from a small language model over a small evidence slice and can be wrong; a clinician must review every answer. Abstentions are escalated. Research / teaching prototype on public exam-style questions - not a medical device.

Category: Department of Computer Science and Engineering / Agent Development

Agent Development Project AG06 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.