Building queries that are needed for categorization of our audio NLP solution. We use categories of conversations to add structure to the data. LLMs (w RAG) seem to do a great first pass.