I build AI infrastructure on my own hardware. There is a two-node GPU cluster (an RTX 3090 and an RTX 5090) that serves local models, an agentic RAG system that retrieves and cites answers against a real 321,000-chunk corpus, and a multi-agent platform of about 32,000 lines that runs a two-model team through a peer-review loop before anything ships. I have built it over the past year and a half — designing it, debugging it hands-on, and folding in what worked from earlier projects, and increasingly using that two-model team itself to write and review code faster.

This work came out of my day job. I test RS-25 flight engines, and I also took on a documentation problem there: a large, scattered pile of technical material that had to be searchable and trustworthy. I cannot share any of that, so I rebuilt the ideas from scratch on public data. That is what these repos are, my own versions of retrieval problems I actually run into at work.

The day job is verify-before-ship. Cryogenic inspections, leak checks, flight-readiness testing, and nothing goes out until it is proven. I hold my own systems to the same standard: config is the source of truth, defaults fail closed, and I label every number measured or estimated instead of just asserting it. I am looking for remote ML-infrastructure or applied-AI work where that kind of discipline is the point.