I'm Ryan Zegalia. 13 years in operations at a mid-size B2C/B2B manufacturer, progressing through warehouse fulfillment, purchasing, e-commerce, marketing, and into the systems and automation I build now. I was the company's first employee, and I built and still operate its integration layer solo.
Below: what I build and run at work, what I run at home, and five industry deep dives. Source code and documentation live in my GitHub portfolio.
Production systems
Systems I build and run at work, shown with the company and product names changed.
Why each piece works the way it does, what broke on the way there, and where the numbers come from: 48 decision records and 9 case studies on GitHub.
Personal production environment
Systems I built and run at home on my own hardware. Nobody else is on call for them, so I've had to make the same calls I make at work about cost, uptime, and what's worth automating. They're on a private network, so each one opens onto screenshots and a walkthrough instead of a live link.
A scoring pipeline over more than 108,000 job postings, built to reject fast and score slow
Six scrapers pull postings into one archive, and a scoring pass ranks each one against the work I actually do. The cheap checks run first, so most postings never reach the model at all.
six sources · reject fast, score slow · local model makes the final call
Read how it works → Self-hosted AIA scheduler and cost ledger for the one GPU my whole home lab shares
Every AI job in the house runs on one card, so I built the scheduler that decides who gets it, the lease that hands the whole card to one job at a time, and the ledger that rebuilds what each program spent.
priority dispatch · time-limited GPU leases · per-program spend ledger
Read how it works → Structured extractionA pipeline that turns community game guides into data a handheld can browse offline
The model reads the guides and picks out the facts, but it isn't allowed to write any of them. Every line on the device is copied from a real guide, and anything it can't ground it skips.
every line verified against its source · abstains rather than guess · runs offline
Read how it works → Computational designAn agent pipeline that rebuilds 3D meshes into editable CAD solids
I start from a downloaded mesh, agents rebuild it into a real editable solid, and I make my changes from there. Every rebuild is checked against the original shape before it's accepted.
mesh in, STEP/3MF out · agents drive the rebuild · live viewer
Read how it works →Industry deep dives
Five industries I don't work in, studied until the operational problems surfaced. The companies are fictional and the data is seeded, but every scenario is worked in full, database and code included. The method is the same one that built the platform above: learn how the industry runs, find the records that disagree, and build the piece that reconciles them.