I'm a software engineer. For the last several years I've worked on the machinery underneath large language models — the routers, schedulers, and storage layers that keep inference cheap and predictable at scale.
Before that I spent a long time writing client software, mostly on Flutter. The two halves of the job inform each other more than I expected: latency budgets, the cost of a wrong abstraction, the value of plain code.
I write here when I have something I'd want to read myself.
Essays · Selected
Why our inference router is mostly a queue
A year of trying to be clever, undone by the surprising sufficiency of priorities, fair shares, and patience.
Notes on a vector store that doesn't lie
Recall curves, the politics of approximate search, and what it took to make our index honest about its own uncertainty.
The shape of a request, from edge to GPU
A walkthrough of one HTTP call — the seventeen hops, the four queues, and the single moment where everything becomes a tensor.
Things I learned writing a custom scheduler
On fairness, starvation, and the unreasonable effectiveness of writing your assumptions down before you write any code.
A short defense of boring databases
Postgres is older than most of the engineers I work with. It's also faster than most of the things we replaced it with.
On reading other people's stack traces
A small love letter to the discipline of sitting with a trace for an hour before you touch the keyboard.
Three years of Flutter, in retrospect
What I'd tell my younger self about pixels, platforms, and the quiet luxury of a render tree you can fit in your head.
Work
I work on inference infrastructure — request routing, GPU scheduling, and the storage shapes that sit alongside them. Most of what I build is invisible to the people it serves, and I prefer it that way.
Earlier in my career I worked on client software, including a long stretch in Flutter; that work taught me to take latency, accessibility, and the small textures of an interface as seriously as anything on the server.
Open to thoughtful conversation, occasional advising, and rarely a new seat. Write to me — I read everything.