Augmented LLM for Text-to-SQL in Practice

Data teams are plagued with constant ad hoc data questions and extract requests from their business counterparts. As a matter of fact, most data teams spend about 30% of their time on these requests while some teams spend as much as 80% of their time. [**Fabi.ai**](https://www.fabi.ai/) leverages LLM to auto-generate answers and create a collaborative workflow for business and data teams to collaborate more efficiently. In this talk, we highlight the challenges of Text-to-SQL in practice, including hallucination, prompt limitation, vagueness of human languages, and distributed business context. We’ll discuss ways to address those challenges and walk through the steps to build a robust trustworthy system. Lessons and learnings from real-world use cases will be shared.