ragtruth

A self-correcting RAG layer that cut hallucination from 35% to 13% on RAGTruth by verifying every claim against the retrieved context.

Privacy-aware enterprise knowledge intelligence. Ask a question over your organisation's documents and get an answer in which every sentence has been checked against a source you are allowed to read — or no answer at all

A diverse 6-LLM judge committee (4 labs, 3 countries) has correlated errors: ~1.9 effective independent judges of 6 (mean pairwise error-correlation ρ≈0.42) on RAGTruth. Provider diversity does not buy error-independence. Verdicts + one-command recompute included.

A reproducible framework for evaluating grounded RAG answers and separating judge errors, benchmark ambiguity, and preprocessing defects.

Empirical evaluation and error analysis of LettuceDetect for RAG hallucination detection using RAGTruth.

Deterministic claim-checking gate for LLM output.

Training-free committee-of-6 LLM judge for RAG hallucination detection: AUROC 0.88 / F1 0.77 on RAGTruth, reproducible at $0 from frozen verdicts — and it surfaces provable label errors in the benchmark.

Evidence-grounded RAG hallucination detection with knowledge graphs and atomic-claim verification.