Querying structured and unstructured data: LLM-first or DB-first?

Is there a way to build data applications on top of information stored both in databases (DBs) and documents in natural language (NL)? We explore the merits and limitations of both Large Language Models (LLMs) and relational databases, questioning whether a LLM-first or a DB-first strategy is more effective to access data in a unified interface. The talk evaluates the role of NL questions and SQL in structured data retrieval and the processing capabilities of the corresponding models. We compare and contrast recent results on these topics and then conclude with an overview of the research challenges in effectively leveraging the combined power of SQL and LLMs.