Du’An Lightfoot is a senior AI engineer at Akamai and the creator of LabEveryday, an educational platform for engineers developing practical technology skills. His work connects enterprise networking and cloud infrastructure with the operational demands of agents and self-hosted AI inference.
On this page
From network engineering to AI infrastructure
An Air Force veteran, Du’An worked in technical support, systems administration, and enterprise networking before joining Cisco as a DevNet content developer engineer. There, he helped infrastructure engineers adopt Python, APIs, Ansible, and network automation through technical writing and developer education.
He created LabEveryday around a practice that gave the platform its name: working on labs daily while learning network engineering and pursuing certifications. His founding account grounds the project in sustained, hands-on learning. That educational work continued as he joined AWS as a senior cloud networking developer advocate and expanded into generative AI and agent development.
Du’An subsequently moved from AWS to Akamai, where his work includes reusable Kubernetes workshop infrastructure for GPU-backed inference. The technical questions extend beyond getting a model to respond: how much memory requests consume, how repeated agent calls affect latency, and what happens when concurrent demand exceeds serving capacity.
Making agents and inference dependable
Inference needs infrastructure-aware engineering. Du’An treats GPU memory bandwidth, key-value cache pressure, batching, tail latency, admission control, and external queues as production concerns. His analysis of inference failure modes examines why an available endpoint can still fail users under load. Understanding prefill, token generation, and cache use helps explain where capacity goes and which changes might improve it.
Dedicated GPU inference gives agent builders operational control. His work with vLLM on Linode Kubernetes Engine connects an OpenAI-compatible model endpoint with monitoring, load testing, security guidance, and agent integration. Running the endpoint brings privacy, latency, rate limits, and token costs into the same engineering decision as hardware capacity and maintenance. His inference teaching with Khaja Omer distinguishes the speed of an individual response from aggregate throughput—measurements that can move differently as concurrency rises.
Network MCP makes infrastructure actionable for agents. Du’An’s open-source network diagnostics server equips agents to investigate connectivity, routing, DNS, and packet behavior. It carries his networking background into tool-enabled AI: the agent gains concrete diagnostic capabilities for examining the systems it depends on.
Agent evaluation and human oversight determine whether automation is dependable. Du’An emphasizes agents that plan, use tools, evaluate outcomes, and adjust their behavior, while retaining deterministic automation for simpler tasks. His browser-agent work with Banjo Obayomi addresses practical limits such as authentication, CAPTCHA challenges, and ambiguous tasks, where verification and human intervention remain necessary.