Nesime Tatbul

Title: Towards Enterprise-Grade Text-to-SQL

Abstract: Recent advances in large language models (LLMs) have accelerated their use across a variety of data tasks. Natural language access to relational databases via text-to-SQL has been a prime example. Public benchmarks have  been instrumental in driving research in this area, but achieving similar results in real enterprise settings with more complex data and query workloads still remains a challenge. This talk presents an overview of our ongoing research in quantifying and closing this gap, by enabling reliable text-to-SQL performance assessment and improvement required for realizing enterprise-grade deployments.