MedAgent-Verify
Trustworthy agentic AI for medical question answering
A multi-step medical question-answering agent built around an explicit verification layer that detects hallucinated claims and triggers safe abstention rather than answering confidently when evidence is insufficient.
The project explores a central question in agentic AI: how do we make a language-model agent reliable enough for high-stakes domains? Instead of trusting a single generation pass, each intermediate reasoning step is checked against retrieved evidence, and the agent is allowed to say “I don’t know.”
Stack: Python, LLM APIs, retrieval-augmented verification