Me

I am a principal AI/ML engineer. Most weeks I am training embedding models, serving LLMs, wiring up retrieval over large collections, and keeping the cloud pieces around them from falling over in production.

I lead small teams on that work too. A prototype is not the hard part. Getting something live — with latency you can actually quote, and a bill that does not surprise anyone — is.

A lot of my work is search and retrieval. I want the system to look at the right evidence, not just sound sure of itself. I have the same itch when the input is not text: time series, video, audio, faces. Language models are useful. They are not the whole job.

I started in biotechnology and mathematics, then spent the last ten years on applied machine learning. Noisy data, a model you can defend, and something other people can run. That is still the job.

On my own time I build tools that watch video, live or recorded, and mark what is happening in the frame. Detection, tracking, small vision-language models. No client brief, which is the point.