Du’An Lightfoot is a senior developer advocate at Akamai, creator of LabEveryday and a builder of AI agents and MCP tools. He teaches the infrastructure work needed to move agents from demonstrations into systems a team can operate.[1][2]
Start with Agents That Own Their Inference →
Links
Du’An Lightfoot: About
Du’An Lightfoot: GitHub 
X · labeveryday Senior Developer Advocate · Akamai
Career
Lightfoot describes a path from the Air Force into enterprise networking, cloud and AI, with work at Cisco, AWS and Akamai. His teaching connects that operational background to code and infrastructure automation. His current public GitHub biography identifies him as a senior developer advocate at Akamai, where he focuses on making cloud networking and automation more accessible. He also publishes LabEveryday videos and practical write-ups.[1][2]
Contributions
Operate the inference layer yourself
In Agents That Own Their Inference, Lightfoot and Khaja Omer teach deployment of a model with vLLM on dedicated GPU infrastructure. The workshop drives concurrent requests through the server so participants can observe batching, cache pressure, throughput and latency. It then changes configuration and measures again. This addresses a different problem from writing an agent prompt: a model that works for one laptop user can behave very differently when several users compete for server memory and generation time.[3][4]
Expose useful tools with MCP
His public repositories include MCP servers for network tasks, PDF reading and Obsidian vaults, along with a Strands agent starter. These are examples of connecting an assistant to a bounded capability or a specific source of information. The collection makes the integration layer visible, rather than treating an agent as a model that somehow gains access to every system around it.[2]
Tools, writing & talks
Writing, research & talks
Start here
Agents That Own Their Inferenceduanlightfoot.comA hands-on dedicated-GPU inference workshop covering vLLM, concurrent load, caching and deployment tuning.
Start with the inference workshop overview and repository if you know containers and basic model serving. Use its concurrent-load measurements to identify one bottleneck, change one setting, and remeasure before scaling a deployment.
Co-instructor with Khaja Omer[3]A hands-on vLLM workshop on running agent inference on dedicated GPUs. It covers streaming measurements, KV caching, FP8 serving and a speculative decoding attempt that backfired, then tunes concurrency and sequence limits by re-measuring each change.
Catalog workOriginal source
Tools
- Akamai inference workshopgithub.com
The code and instructions supporting the dedicated-GPU agent-inference workshop.