Hands-on notes from real infrastructure and AI work — what worked, what broke, and what it means for the businesses and teams weighing the same decisions.
Skipping Akamai's managed Kubernetes to build a cluster from scratch with kubeadm, MetalLB, and a hand-rolled NFS server — the strategy and the build. Part 1 of 2.
TP-Link's Omada SDN platform has no official Ansible support and no way to write VLANs through its Python client. Building the missing piece meant reverse-engineering an undocumented API — and turning a client's network forklift into a reusable Ansible collection.
A commercial security camera project that grew from six cameras to fourteen — and the network architecture, storage math, and vendor decisions that came with it.
A growing body of research on AI and cognition — MIT's 'Your Brain on ChatGPT' study, classroom findings from Oregon State and Stanford, and the case for using AI deliberately instead of by default.
A hands-on evaluation of an AI agent with real terminal access to a local Kubernetes and Docker environment — and why stronger reasoning, not more tools, is what actually prevents dangerous mistakes.
AI appears to be consuming the specialized skills that built it. That looks like destruction. It might be the collapse of the barrier between having an idea and building it.
A hands-on walkthrough of building a retrieval-augmented generation pipeline over 510 real legal contracts — chunking strategy, vector search, an open model served with vLLM, and the environment realities the tutorials leave out.
A field guide to what actually goes wrong when you self-host an open-source AI model — CUDA version traps, mystery disk quotas, and a quantization result that surprised me. Notes from a hands-on infrastructure lab.
Applewhite