Debugging LLM Models

I built a small CLI tool that debugs bad LLM outputs by treating another LLM as a critic. It takes a broken prompt-response pair, diagnoses the failure, and rew