Schedule

All times in America/Los Angeles.

6 sessions

10:00 AM – 10:45 AM PDT
AI EngineeringMain Stage45-minute talk

Most RAG demos fall over the moment real users show up. We rebuilt our retrieval stack three times; this talk is the version that stuck, including the chunking decisions we regret and the reranker that finally paid for itself.

Priya Raman — Staff ML Engineer, Northwind Labs

11:00 AM – 11:30 AM PDT
AI EngineeringCommunity Hall30-minute talk

A tour of the batching, caching, and routing changes that took our monthly bill from eye-watering to unremarkable — with the measurements that proved quality held.

Tom Beckett — Principal Engineer, Halyard

11:00 AM – 11:45 AM PDT
Evals & ReliabilityMain Stage45-minute talk

Everyone writes an eval suite. Almost nobody maintains one. A practical pattern for evals that stay green, stay fast, and catch the regressions that matter.

Hannah Kim — Research Engineer, Blue Harbor AI