A self-correcting RAG layer that cut hallucination from 35% to 13% on RAGTruth by verifying every claim against the retrieved context.
fact-checking nli rag self-correction llm faithfulness retrieval-augmented-generation llm-evaluation hallucination-detection ragtruth
-
Updated
Jun 21, 2026 - Python