AI Infrastructure Lab
Local models, connected tools, and lessons from the workbench.
Personal lab / ongoingDigital systems consultant
FACKO — Systems. Automation. Experiments.
Systems.
Automation. Experiments.
I build, break, and improve digital systems.
Frankfurt Córdoba
Selected work
Inside the lab
Personal projects / always learning
Home Automation
One room-first dashboard. A home that is easier to use.
Personal home / ongoing
Homelab / Media Stack
Self-hosted services, connected carefully and kept running.
Personal infrastructure / ongoingOriginal pixel artwork; project details document the actual work.
How I work
Curiosity in. Clarity out.
Understand
A clear problem and a way to judge the result.
Map
A small system map. Clear responsibilities.
Build
Something usable, with room to learn.
Verify
Evidence, a useful handover, and the next iteration.
A closer look at the process
Understand the real problem.
Start with the person using the system. Look at the friction, the constraints, and what a useful result would actually look like.
Map the moving parts.
Trace the flow of information. Make ownership, access, dependencies, and failure modes explicit before adding another tool.
Build a small working version.
Connect the essential pieces first. Automate repeatable work, keep changes reversible, and use AI where its output can be checked.
Verify. Document. Improve.
Test actual behavior, including when something is unavailable. Write down decisions, recovery steps, and what still needs work.
Start with the room.
In Home Assistant, I brought scattered controls into one room-oriented dashboard. The starting point was how the home is used.
Home Automation ↗Give each layer a job.
In the AI lab, the interface, model gateway, and inference service have separate responsibilities and explicit tool boundaries.
AI Infrastructure Lab ↗Test the service, not the label.
A VM that looked “running” had stopped answering. The incident changed what I check: a real request, not just a green status.
Recovery field note ↗From the workbench
Field Notes
GPU passthrough broke again
The VM looked alive. The inference service was not. A reset problem that needed a physical recovery.
One dashboard is enough
The home interface became simpler when there was one obvious place to start.
Quiet systems
The best infrastructure is the kind nobody notices.