Loved the product. Had been struggling with my n8n agent which writes shitty queries, irrespective of how many times the schema is provided. Question: What are you using for evals? How are you making sure of the performance of your AI copilot?