San Francisco in May means conference season. From May 12 to 14, AI Council brought data and AI practitioners to the Marriott Marquis: analytics engineers, ML platform leads, founders, infrastructure architects, and investors gathered for three days of conversation about what's actually working in production right now.
We set up shop in the exhibitor hall with the full booth crew.
The signage behind us read "More than a feature store, the data platform for AI + ML," which turned out to be a useful conversation starter. The phrase "feature store" still means different things to different teams, and a lot of the booth traffic came in with a version of the same question: What does it actually take to make a model react to live data without breaking everything upstream?
Our forward-deployed engineers spent most of the three days working through that question, one whiteboard at a time. The format was the same one we've leaned into all year. No slides, no pitch deck. An engineer pulls up a chair and walks through the architecture with whoever sat down. Rishi anchored the booth with live demos, Jackson kept the calendar moving, and our co-founders rotated through for executive conversations between sessions.
A few patterns came through clearly across the conversations.
The room is past the "should we use AI" debate. Teams have models in production. What they're hitting now is the messier second wave: features that look fine in training and fall apart at inference, pipelines that can't keep up with real traffic, the same Python logic rewritten three times across notebooks, training jobs, and serving systems. The question has shifted to how teams operationalize ML without a six-month platform project.
That gap, between the model that works in the notebook and the model that works in production, is where Chalk lives. AI Council was a useful reminder of just how many teams are arriving at the same problem from different angles. Recommendation systems, fraud and risk, payment authorization, underwriting. Different surface areas, same underlying need for fresh data, computed correctly, served fast, with a warehouse that does its job and a layer above it that does the rest. It's the same problem teams like Whatnot, Mission Lane, and MoneyLion came to us with, just told back to us in new accents.
We talked to a lot of those teams this week. We'll be talking to more of them at the next stop.
What's next
Two weeks from now, we head to Snowflake Summit, June 1 to 4 at Moscone, for what's shaping up to be one of our biggest events of the year. Marc and AJ Balance, CPO at Grindr, are taking the main stage to walk through the real-time infrastructure behind one of the largest consumer AI apps in the world. Elliot is leading a technical breakout on a computation-first architecture for real-time ML feature stores. The full Chalk team will be at Booth #2611 all four days.
And if you've been around San Francisco the past few weeks, you've probably seen us already. The Chalk bus is on the streets, running its loop through the city. Why a bus? You'll find out June 1 at Summit. Head to chalk.ai/speed to get on the list.
See you in June.
