verbatim-rag and RAGGuard
These are complementary tools: Verbatim-RAG prevents hallucinations through grounded generation with exact citation extraction, while RAGGuard detects hallucinations post-generation through faithfulness scoring, allowing them to be used together in a defense-in-depth approach to hallucination mitigation.
About verbatim-rag
KRLabsOrg/verbatim-rag
Hallucination-prevention RAG system with verbatim span extraction. Ensures all generated content is grounded in source documents with exact citations.
Implements pluggable span extractors—either LLM-based or fine-tuned ModernBERT models—to extract verbatim passages from retrieved documents rather than allowing free-form generation. The entire pipeline can run without LLMs and on CPU-only hardware using SPLADE sparse embeddings and Milvus vector storage, with both core extraction logic and a full web interface (FastAPI + React) included.
About RAGGuard
MukundaKatta/RAGGuard
RAG hallucination detection — verify LLM responses are grounded in source documents with faithfulness scoring
Related comparisons
Scores updated daily from GitHub, PyPI, and npm data. How scores work