Agents That Own Their Inference — Du'an Lightfoot & Khaja Omer, Akamai Technologies

Why watch

A hands-on workshop on optimizing LLM inference for AI agents using vLLM on Kubernetes, covering speculative decoding, FP8 quantization, and throughput/latency tradeoffs. Participants work with Qwen models to measure memory budgets, KV caching, and concurrency limits. Designed for ML engineers and platform engineers building production inference systems.