How I Built a Hallucination Detector for RAG Pipelines in Python

The author built HallucinationBench, a Python tool that uses GPT-4o-mini as a structured judge to detect hallucinations in RAG pipeline outputs. It breaks LLM responses into individual claims, classifies each as grounded or hallucinated against the retrieved context, and returns a faithfulness score and verdict. The approach requires no embeddings or vector databases — just a single API call with JSON mode and temperature 0 for deterministic results.