Most of what I've made public is research code from the PhD; the exception is Tendrel, which I use every day.
A Claude Code plugin that keeps a research graph inside your repo: what's been tested, what held up, what's blocked, and what falls over when a result changes. I built it because agents are sharp within a session and forget everything between them, so I kept making decisions on a stale picture of my own work. Plain markdown, no database, no daemon. Maintained and MIT licensed, with a small user base.
The language model from my PhD: a general model adapted to veterinary clinical notes with almost no labelled data to learn from, released so other people could build on it. It is the repo people email me about most.
Reconstructs a patient's journey, a temporal phenotype, from observations scattered across free-text clinical records over time.
Maps veterinary inventory items and labels to their antimicrobial ingredient. The unglamorous step that made the antimicrobial usage work possible.
Supporting code for the antimicrobial prescribing papers.
github.com/havocy28/prescription-text-analyzer · github.com/havocy28/am_usage
The rest is on GitHub.