Write code to detect and handle hallucinations in LLM outputs

Implement a post-generation check that verifies an answer against its source context, flags unsupported claims, and decides whether to return, regenerate or refuse.