Why it matters

Hallucinations damage trust. Detection shapes reliability.

Advertisement

The architecture

Verify: check output against retrieved context.

Fact-check: external sources.

Hallucination detectionOutputfrom LLMVerifyagainst contextFlag or fixif unsupportedCombine: NLI models + citation checking + external verification
Detection pipeline.
Advertisement

How it works end to end

NLI (natural language inference): does output follow from context?

Citation check: are cited sources real?

External verifier: query fact API.