[SV] AI meetup for GenAI, LLMs, and Agent

​Welcome to the AI meetup in Silicon Valley. Join us for deep dive tech talks on AI, GenAI, LLMs and Agent, hands-on experiences on code labs, workshops, and networking with speakers and fellow developers.

​Agenda:
- 5:30pm~6:00pm: Checkin, food and networking
- 6:00pm~6:10pm: Welcome, Community update
- 6:10pm~8:00pm: Tech talks and Q&A
- 8:00pm~8:30pm: Open discussion, Mixer and Closing.

​Tech Talk: Stop Writing Text-to-SQL: Give the LLM the Job It's Actually Good At 
Speaker: Ashish Shubham, Engineering Fellow at ThoughtSpot 
Abstract: Most agentic analytics projects start by asking an LLM to write SQL. It demos well and fails in production. The problem isn't broken syntax; it's queries that run, return a number, and are quietly wrong. Fan traps. Unvalidated join paths. Same question, different answer.
We'll present the right way to build instead: a semantic/context layer where metrics, join paths, and permissions are defined once and enforced deterministically, exposed to the agent as a narrow tool surface rather than a SQL field. We'll close with a real working solution built on this architecture and its trade-offs.

​Tech Talk: Stop Before the Agent Writes Code: Triage Is the Real Bottleneck
Speaker: Jakob Prince, PromptQL 
Abstract: The industry has largely solved "can the agent write the code." It has not solved "was the agent handed the right problem." A coding agent given a badly specified bug will produce a clean, well-tested, fully green fix to the wrong thing, and every downstream validation step will pass it. I'll walk through what happens in the ten minutes before the agent gets the task: pulling in the context that's scattered across people and systems, letting a teammate correct a wrong assumption in real time, and only then handing off a task that's specified well enough to be worth validating.

​Tech Talk: The Stack Behind No-Code Agents That Don't Suck
Speaker: Sam Bowerman at ThoughtSpot 
Abstract: You can already build agents. Your HR, Legal, and Finance teams can't, so they file tickets and you become the bottleneck. Making agents usable by non-engineers is harder than it sounds: no code, but they must still be deterministic, governed, and shared. That means a context layer instead of prompts, permissions that survive delegation, and multiplayer state when five people build on one agent.

​Speakers/Topics:

​Stay tuned as we are updating speakers and schedules.

​If you have a keen interest in speaking to our community, we invite you to submit topics for consideration: Submit Topics