Built a self-verifying RAG pipeline that refuses to hallucinate confidently — uses an independent reranker (not the generating LLM) to judge answer quality, requires 4-source consensus, and scales responses across a 5-tier confidence system down to hard refusal.
Like this project
Posted Sep 3, 2026
Built a self-verifying RAG pipeline that refuses to hallucinate confidently — uses an independent reranker (not the generating LLM) to judge answer quality, ...