PrismShine catches AI hallucinations by checking evidence, not just text

PrismShine, an open-source tool from InsightITS, detects AI hallucinations by verifying answers against source evidence before and after generation. In benchmarks, it achieved 0.831 F1 on QA tasks and perfect 1.000 on number accuracy, outperforming Vectara's HHEM-2.1-Open. The tool runs on CPU with no LLM calls by default, offering auditable gates for debugging.