A short version.

I lead a small research group inside a federally-funded research and development center, where we build AI systems against the kind of problems that don't show up in a benchmark — messy schemas, regulated data, stakeholders who don't yet know what they want. I've been doing this for about a decade, in one form or another, and the problems that interest me now are almost all about the middle layer: the part between the model and the business where most of the actual engineering happens.

Before the current role I did research at Harvard and spent time in a couple of earlier analytics and platform roles — the details are in the resume, which I'm happy to send. The career arc is less interesting than the through-line: I've been drawn, consistently, to systems where a smart model meets a hard data environment, and the only path forward is to get the plumbing right.

The writing on this site is the long-form version of what I'd say if you caught me at a conference. The sandbox — PokeScan, the MCP experiments, whatever's next — is where I test patterns before I'd put them in a real deployment. Nothing here is a product; it's all notes toward a craft.

If any of this overlaps with something you're working on, reach out. I'm happy to talk.