The Homelab
Proxmox, k3s, four VLANs, and local AI. A real production network that happens to live in my house.
The problem
I wanted real hands-on infrastructure experience, not just tutorials. Reading about VLANs and Kubernetes only gets you so far — you don’t find the sharp edges until something is actually running and something actually breaks.
What I built
A homelab that’s grown into the whole stack:
- One cluster: 2 Proxmox hosts, a 4-node k3s cluster, and 5 Raspberry Pi 5s
- Segmented VLANs: trusted, IoT, DMZ for public-facing services, and guest
- A Tailscale subnet router for remote access
- TrueNAS for backups
- Home Assistant running the house
- A self-hosted media stack
- Local AI on an RTX 5070 Ti (Ollama, faster-whisper)
- Monitoring with Beszel and Uptime Kuma — 55 monitors, alerting to Telegram
Stack
Proxmox, k3s, TrueNAS, Tailscale, Home Assistant, Ollama, faster-whisper, Beszel, Uptime Kuma, Telegram.
What broke and how I fixed it
A voice-assistant TV integration kept taking over the screen mid-show to show an announcement. The fix was routing announcements through a self-hosted dashboard page instead of casting to the TV directly — the dashboard already had a view, so announcements just update it instead of hijacking the input.
Separately, a monitoring checker kept marking healthy hypervisors as down. Turned out they were answering HEAD requests with a 501 Not Implemented, and the checker read that as an outage. Falling back to a GET request fixed it.
What I’d do next
Actually test off-site backup restores instead of trusting that backups exist, and write runbooks so “how do I fix this” doesn’t live only in my head.
Like this one
The resume has the rest of the story.
See resume →