Things I have built.

Model-serving infrastructure and the products around it, then a few public repositories where the code itself is the write-up.

The API layer and control plane for a scalable language-model inference server.

A platform for putting heavy language models onto GPU virtual machines — serverless or clustered, depending on the load.

A retrieval-augmented chatbot with an admin console for its knowledge base, and an embeddable widget that drops into any site with one script tag.

Interfaces that let non-technical users run Google Cloud operations — Dataproc jobs, BigQuery tables, GCS datasets — without touching the console.

A front-end rebuild of a loan distribution system, aimed squarely at the errors users kept hitting.

In the open

Work with the source attached — smaller, but you can read every line of it.

There is more on GitHub — smaller interface experiments, and the scratch work that never turned into anything worth writing up.