Personal lab / ongoing
AI Infrastructure Lab
Local models, a private knowledge base, and a sandbox with read-only tools. A home-scale AI system, including the parts that break.
Projects, notes & other loose ends.
Projects and notes
A home for my projects, small discoveries, and things I'm still figuring out.
Look around
01 / Things I've been building
The project shelf
Scroll sideways to see more projects.
Personal infrastructure / ongoing
Homelab / Media Stack
Virtual machines, containers, segmented networking, and a self-hosted media stack.
Personal home / ongoing
Home Automation
Home Assistant, Hue, Zigbee sensors, and a dashboard that starts with the room.
Archive / research outline
Private Monitoring Node
A notebook thread about useful health checks, private dashboards, and the difference between “running” and “working”.
Archive / research outline
Digital Identity Archive
A personal archive thread: old accounts, neglected exports, and the question of what remains when a service disappears.
Personal lab / ongoing
Model Router
LiteLLM as the one door to the models, a local search over my own notes, and a small keyword router so the prompt stops carrying tools nobody asked for.
02 / A thought, before I forget
The notebook
workbench
I only learned this to spend fewer tokens
I did not want a proper AI stack. I wanted the small local model to stop swallowing a huge prompt for every question.
incident
GPU passthrough broke again
The VM still showed as running, but inference had stopped. Restarting the host brought it back.
workbench
One dashboard is enough
I made one room-oriented Home Assistant dashboard the starting point and moved the older ones aside.
03 / Behind the work
How I work
From the first question to the failure test.
Explore the processClose the process
03 / Behind the work
How I work
From the first question to the failure test.
Explore the processClose the processSystems
How I work
The tools change. I still want to know what depends on what, and how to get it running again.
Understand the problem
I start with whoever uses the system: what gets in the way, and what would make the work easier?
Map the dependencies
I trace where the information goes and which service owns each part. This usually explains more than adding another tool.
Try a small version
I connect the essentials first and keep a way back. I use AI where I can check what it produces.
Check the failure case
I test what happens when a service goes away, then write down how to recover it and what still needs work.
From my own projects
Home Automation
I put the scattered Home Assistant controls into one dashboard, organised by room.
AI Infrastructure Lab
The chat interface, model gateway and inference service each have their own job. Tool access has explicit limits.
GPU passthrough broke again
The VM said it was running but stopped answering. Now I check it with an inference request.
I leave recovery steps and open questions in the documentation, so the next person has somewhere to start.
Get in touch