Coalesce 2024: Building a hallucination-proof SQL AI with MetricFlow and LLMs
Have you ever wanted to build an AI chatbot to query your data platform? Transforming a business user's natural language question into a valid SQL query that answers their question has huge potential to unlock improved self-service analytics.
Have you ever wanted to build an AI chatbot to query your data platform? Transforming a business user's natural language question into a valid SQL query that answers their question has huge potential to unlock improved self-service analytics. However, even state-of-the-art LLMs are prone to hallucination, which makes it hard to generate SQL queries that are both conceptually and syntactically correct. In this talk, you'll learn how the team at M1 Finance tackled this problem using MetricFlow’s rich semantic modeling along with LLM-based tool invocation. This approach enables them to reliably produce valid SQL queries from natural language questions, avoiding issues of hallucination or incorrect metrics.
Speakers:
Kelly Wolinetz, Senior Data Engineer @ M1 Finance
Brady Dauzat, Machine Learning Engineer @ M1