BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//sf26-agenda//EN
CALSCALE:GREGORIAN
METHOD:PUBLISH
BEGIN:VEVENT
UID:SESS-35@sf26-agenda
DTSTAMP:20261020T151709Z
DTSTART:20261105T213000Z
DTEND:20261105T221500Z
SUMMARY:Teaching datalakes to speak human with LLMs and MCP
LOCATION:Room 2006
DESCRIPTION:Olena Kutsenko\n\nImagine asking your data storage a question i
 n plain English - and getting the right answer back\, instantly. Thanks to
  advances in large language models (LLMs)\, this is now a practical realit
 y. Natural Language to SQL (NL2SQL) systems are making data more accessibl
 e than ever\, helping teams move faster without writing a single query by 
 hand. In this talk\, we’ll walk through the key building blocks that mak
 e it possible for LLMs to "talk" to databases. We’ll start with natural 
 language: how NL2SQL systems understand what the user is asking\, map ques
 tions to the right parts of a database\, and generate executable SQL. Of c
 ourse\, this is easier said than done. Natural language is full of ambigui
 ty\, and many databases have complex schemas\, tricky joins\, and domain-s
 pecific terms. But despite these challenges\, benchmarks like Spider and B
 IRD show just how far we've come in the past decade. Next\, we’ll introd
 uce the Model Context Protocol (MCP) - a way to give LLMs access to metada
 ta\, table relationships\, and tools for query execution. Instead of guess
 ing\, the model can reason step-by-step using chain-of-thought\, consult t
 he schema\, and run sub-queries to reach the right result. Whether you're 
 an engineer building LLM-powered interfaces or a data leader exploring sel
 f-serve analytics\, this session will give you a clear view of how natural
  language is reshaping the way we interact with data and how to start usin
 g it in your stack today.
URL:https://sf26.rmoff.net/s/teaching-datalakes-to-speak-human-with-llms-an
 d-mcp-35
END:VEVENT
END:VCALENDAR