An opinionated case for why infrastructure engineers should learn Python as an engineering tool — automation, APIs and glue, not becoming a developer.
An opinionated, experience-grounded look at why most AI projects fail — data quality, governance, hype, ownership, security and operational readiness — and what success really takes.
Why the first conversation matters more than any later technical decision, and the handful of questions I ask in discovery that quietly determine whether a project succeeds or fails.
Why Citrix session density is bounded by profile container IOPS long before CPU saturates — the sizing trap I keep watching projects fall into, and the rule of thumb I size from instead.
Why the most useful idea in automation is making a script safe to run twice — and how designing for re-runs instead of one perfect pass quietly fixed most of my reliability problems.
How n8n became the connective tissue between my notes, my AI assistant and my infrastructure — and the silently-swallowed webhook payload that taught me to distrust the happy path.
What embeddings actually are, why they turn meaning into geometry, and why every retrieval system — including mine — lives or dies on getting them right.
How I actually choose which local model to run — the trade-off between size, quantisation and VRAM, and why the biggest model my GPU can hold is almost never the right answer.
How a single GPU and a native Ubuntu install became the box that runs my local models — and why I gave up on virtualising it and went straight to bare metal.
Moving the homelab from hand-managed proxy rules to a single Caddyfile with automatic HTTPS — and the DNS-challenge gotcha that cost me an evening before it all clicked.