Comment by eka1

Did you validate this by running a A/B test? Main question is were you able to classify back into your known categories correctly all the time, or did the errors compound from the llm hallucination plus embedding search