PriHA: A RAG-Enhanced LLM Framework for Primary Healthcare Assistant in Hong Kong

A collection of recent RAG/LLM research papers reporting advances and failure modes relevant to trustworthy, domain-governed retrieval. Key security-relevant findings: the "Controlling Authority Retrieval (CAR)" paper formalizes retrieval of authoritative, superseding documents (e.g., security advisories, legal rulings, FDA records) and shows dense retrieval often misses active frontiers (security advisories Dense TCA@5=0.270 vs Two-Stage=0.975) and produces incorrect "not patched" claims in 39% of queries where patches exist (reduced to 16% with a Two-Stage approach). FRESCO exposes a re-rank