Shane Connelly

Background

Wrote Vectara’s introduction of the Hallucination Evaluation Model (HEM) and its public Hallucination Leaderboard, which scores whether a model “only uses the data provided to it” when summarising retrieved results.

Articles in This Vault

• Measuring Hallucinations in RAG Systems

Key Contributions

• Made groundedness a comparable, published metric across models rather than a per-project judgement
• Surfaced the correlation between summary length and hallucination rate

Concepts

• Hallucination Detection · RAG · LLM as Judge · Search Evaluation