Hi, I'm Jason.
I ship code that
teams run in prod.

I treat AI engineering as a systems problem, not a prompt problem.

The interesting work is making LLMs reliable: orchestrating multiple agents, designing the control loops and state around them, and building the infrastructure so the whole thing holds up under real load. I run a heavily customized multi-session agent workflow over a FastAPI/SQLite control dashboard — the kind of tooling that turns a clever demo into something you can actually depend on.

That instinct isn't incidental — it's a master's in systems design with an AI focus, plus seven years shipping production software: machine learning, backend engineering (TypeScript/NestJs, Java/Spring Boot), and now production LLM systems. The same instinct shows up in open source, where I contribute fixes to the infrastructure other people build on.

Two agents you can click.

Not screenshots — deployed and running. Open either one and watch the system work. This is the difference between saying "I build agentic systems" and handing you two of them, live.

From ML to systems, and back to AI.

Seven years, with AI at both ends. I started as a machine-learning engineer, went deep on backend systems, then came back to AI — building with LLMs in production back in 2024, before "agent" was a job title.

Merged, not "side projects."

Real code, reviewed and merged into projects teams run in production — from Apache infrastructure to the LLM tooling the AI ecosystem builds on. Every card links to the PR; all public and verifiable.

Plus a few more merged across these repos.

Shipped solo — idea to store.

A spread of consumer apps, on-device CV/ML experiments, and dev tooling. I design, build, and ship them myself, often coordinating parallel agent workflows to move faster.