Skip to content

BRK422 Breakout Thu 19 Nov 17:15 PST

Run open-weight models, post-training, and agents on AKS

Lachie Evenson, Jorge Palma, Jeremy Cohen, Shanshan Song

Session description

Microsoft's catalogue description

Serving open-weight models, post-training on new data, and running agents that take action each place different demands on GPUs, isolation, and cost. This session shows how Azure Kubernetes Service (AKS) and open source give platform teams control across all three: scaling GPU capacity, serving models, and isolating agents, with Ray for distributed post-training and MCP for connecting agents to tools. Hear how a customer runs AI in production on AKS and leave with patterns for your own tenant.

← All sessions