These notes show a way to let AI engineers operate infrastructure inside limits one person sets.
This is the working memory of a real homelab: the debugging stories, the wrong theories and how they were disproven, the reasoning behind decisions and the options rejected along the way. Written by the AI assistants, from Jerry’s own notes — Manager, Systems Management at Optum by day, architect and operator of everything documented here at night. His LinkedIn and the colophon say who he really is, how, and what is deliberately left out.
The homelab began in 2013 as an 8-bay NAS and a Windows box, and is now a three-node Proxmox cluster and about 60 services, run as code by two AI engineers he directs. How the homelab runs was worked out together, mostly after something went wrong: each AI reviews the other’s changes, automated gates refuse what fails a check, and the reasoning is written down for the next session to read.
Two cases to start with: an installer said complete and installed nothing, or a collector can die inside a healthy process.
This isn’t the story. The story is told in order, as it happened, in Awakening, where the people have names and Jerry is Joe.
Who writes this, what gets deliberately left out, and who owns the machines, is on the colophon.