BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//sf26-agenda//EN
CALSCALE:GREGORIAN
METHOD:PUBLISH
BEGIN:VEVENT
UID:SESS-35@sf26-agenda
DTSTAMP:20261103T101708Z
DTSTART:20261105T213000Z
DTEND:20261105T221500Z
SUMMARY:Teaching the Streamhouse to Speak Human with LLMs and MCP
LOCATION:Room 2006
DESCRIPTION:Olena Kutsenko\n\nImagine asking "What changed in the last few 
 minutes\, and who or what is affected?" and getting an answer based on liv
 e data\, with events connected\, context built in real time\, and structur
 ed sources you can trust. This is where the Real-time Context Engine meets
  the Streamhouse: turning constantly changing data into context that LLM-p
 owered agents can explore\, query\, and reason over. But it takes more tha
 n just plugging an LLM into a data source. We'll follow events from where 
 they're created all the way into the Streamhouse\, where schemas help us m
 ake sense of the data and manage changes. We'll explore when to append eve
 nts to keep the full history and when to upsert records to maintain the la
 test state. Then we'll see how stream processing turns all of this into qu
 eryable tables\, powered by a low-latency query engine. Next\, we'll see w
 hat happens when a user asks a question. An LLM-powered agent can discover
  available data\, explore schemas and metadata\, and run bounded queries u
 sing tools like Model Context Protocol. We'll break down how it all fits t
 ogether—and what it takes to make it work reliably in production: handli
 ng late events\, access controls\, query limits\, failures\, freshness\, a
 nd making sure people can verify the answers. By the end\, we'll have conn
 ected the dots from streaming events to answers people can actually use\, 
 with the context and controls to trust them.
URL:https://sf26.rmoff.net/s/teaching-the-streamhouse-to-speak-human-with-l
 lms-and-mcp-35
END:VEVENT
END:VCALENDAR