Ryan Gattis

I'm a software engineer in Seattle working on real-time 3D systems — renderers, GPU compute, and the measurement infrastructure that makes performance work something other than guesswork.

My name is Matthew Ryan Gattis and I go by Ryan, which is why my GitHub account reads matthewgattis. Same person.

Where I've worked

Eight and a half years at E4D Technologies building CAD/CAM software for restorative dentistry — taking 3D intraoral scan data through design and into manufacturing. It is a good domain for learning rigor, because geometry errors do not stay abstract. A miscomputed sprue angle produces a restoration that will not seat in someone's mouth.

After that, a contract at Microsoft on HoloLens 2 and IVAS as an XR Software Design Engineer, where most of what I built was diagnostic tooling: latency and dead reckoning instrumentation, head tracking metrics surfaced to the application layer, and runtime anchor validation that made spatial drift visible instead of anecdotal.

How I work

The through-line in both jobs, and in most of what I build on my own time, is that I would rather measure than speculate. On simulife-rs I built Chrome-trace instrumentation into the server and viewer before optimizing anything, then let the captures set the agenda.

The best argument for doing it that way was a find I would not have reached otherwise: two async tasks that had quietly ended up on the same worker thread, running one after another instead of in parallel. No function was slow, nothing in the code looked wrong, and a profile sorted by self time says nothing about it. On a timeline it is obvious. Separating them nearly doubled server throughput.

Some of my recent projects had their implementation generated rather than hand-written, and each of those says so at the top of its page. On them I designed the architecture, chose the approach, and did the analysis; the code was written by a model. The design documents in those repositories are the clearest record of which parts were mine.

That label means something specific, so it is worth saying what it does not mean. It does not mark every project I used a model on at all — I use them the way I use any other tool, and marking that everywhere would make the label meaningless on the projects where it actually matters. It marks the projects where I did not write the code. Everything unmarked I wrote, and I can account for any part of it.

What I'm looking for

Graphics, performance, or systems engineering — ideally work where the correctness bar is real and the feedback loop runs through measurement rather than opinion.

The best way to reach me is ryangattis@outlook.com.