Deploy DB agent with Agentic Memory on Databricks
Join this event
About this event
Most "AI agent memory" demos assume one agent, one database, one trust boundary. Real enterprises don't work that way: your OLTP system lives on SQL Server, your analytics warehouse is on Databricks or Snowflake, and they sit in different security zones for good reason. The moment you try to give agents on both sides shared context, you've either built a data leak or given up on memory entirely.
In this hands-on session, we walk through a redaction-first alternative built and tested against real infrastructure: instead of agents sharing raw questions, SQL, or result rows, each interaction is passed through a second LLM call that produces a sanitized, structured summary, referencing entities by ID and type, never literal values, before it's written to a shared store. Other agents, running on completely different platforms, retrieve that context without ever seeing the underlying data.
We'll build this live: two agents, two platforms, one shared memory store, zero raw data crossing the boundary. AWS S3 Vectors as the backend, verified end-to-end against real AWS. And the exact bug this architecture caught in production testing, a silent embedding truncation, found and fixed during real work, not staged for the demo.
No slides. Real code, live in a notebook, shared with every registrant afterward.
You'll leave knowing:
- Why session-recall memory (what most "agentic memory" content means) doesn't survive contact with enterprise security boundaries
- The summarize-then-store pattern, and why redaction has to be a second LLM call, not a filter
- How to evaluate whether a memory architecture is safe to deploy across more than one trust boundary
- What broke in real testing, and what the fix looked like
Who should come: data and analytics engineers, platform and security leads, and anyone building on the open-source db-agent project who wants to see the maintainer walk through what's newest.
About the Instructor
Chandan Kumar
- Founder of BeCloudReady (Databricks Registered Partner)
- Organizer of TorontoAI (10K+ members).
- Author of the open-source db-agent project (github.com/db-agent/db-agent), presented at the AAAI-25 workshop on AI agents.
Live
- Live on Microsoft Teams.
- Recording sent to attendees within 48 hours.
- Saturday, Aug 20 — 5 PM EDT.
Build something real. Not just prompts.
URLs
Want to find AI jobs?
Join thousands of AI professionals finding their next opportunity