$ cat ~/blog/*.mdx
Software, homelab, and AI experiments, usually written after something breaks.
How I run Qwen with Ollama: on a MacBook for trying models, and on an always-on Mac mini behind Tailscale as a private model server for a real app. The headless-Mac gotchas, and the cold-load bug that made every quiz question take 158 seconds.
Two days of overdue homelab maintenance (251 pending packages, a Pi-hole upgrade, community regex rules, a broken Grafana Alloy install), done by Claude Code over SSH, with sudo kept firmly in human hands.
How I turned a headless mini PC and an existing AI agent into a hands-free voice assistant for the car, and the two bugs that taught me the most.
A GrapheneOS phone stopped showing contact names in Messages. The permission screen said Allow. The OS quietly said no. Debugging it over adb — and what it took to back the phone up before leaving for stock Android.
Questions or corrections? Find me on LinkedIn.