events

Faster AI: GPU Rooflines and Agent Learning

An online session from the AI Performance Engineering Meetup (London–Cambridge), hosted by Chris Fregly and Antje Barth, with short technical talks on making AI systems run faster. Rajat Dandekar of Vizuara explains how concurrent users cha

In inglese

Starts

2026-10-19T17:00:00+00:00

Organizer

AI Performance Engineering Meetup (London - Cambridge)

Location

Online event

An online session from the AI Performance Engineering Meetup (London–Cambridge), hosted by Chris Fregly and Antje Barth, with short technical talks on making AI systems run faster. Rajat Dandekar of Vizuara explains how concurrent users change GPU utilization, linking vLLM measurements on an H100 to the roofline model. Rustem Feyzkhanov of Snorkel AI then describes a loop from production agent traces to reproducible simulations, verifiable rewards and model improvement through fine-tuning or reinforcement learning, and when changing prompts or tools works better than updating weights.