Inference on Kubernetes / Building a Multi-Avatar System with Agentic AI
Hosted by Los Angeles DevOps
Thursday, November 19, 2026 · 6:00pm · O'Briens Irish Pub · hosted by Los Angeles DevOps
about this room
Los Angeles DevOps takes over the covered patio at O'Briens Irish Pub in Santa Monica for two talks about running AI on real infrastructure. Hrittik Roy of vCluster Labs, a CNCF Ambassador, walks through what changes when a local LLM moves onto Kubernetes under load: prefill and decode, the KV cache, batching, GPU scheduling and when disaggregated inference earns its complexity, using vLLM and NVIDIA Dynamo as examples.
Arsen Yeremin, CTO at Tygher AI, then shows a system where several photoreal 3D avatars each run their own agent on a self-hosted five GPU cluster, with live demos covering turn-taking, Gaussian Splat characters and mixed GPU workloads.
Doors at 6pm, intros at 6:30pm, talks until 8pm, then an hour of networking. Meters nearby expire at 6pm and the patio gets cold, so bring a jacket.
two practitioners showing the actual plumbing behind LLM inference and multi-agent systems, in a room small enough to ask them about it afterwards.