// data scientist · ai engineering
Agents, evals, and the tradeoffs nobody puts in the README.
Part 5 of building a wardrobe AI: why per-item feedback hits a ceiling, and the weekly LangGraph that distills preference rules from my whole history, designed around the fact that it might be wrong.
Part 4 of building a wardrobe AI: turning the app into a trip packer, and using LangGraph to let it branch off and shop when it spots something I don't own.
Part 3 of building a wardrobe AI: how the app learns my taste from a single thumbs-down, why that's harder than it sounds, and the difference between a loop that self-corrects and a bias that doesn't.
Part 2 of building a wardrobe AI: the deterministic pipeline that curates which clothes the model gets to consider, and the prompt that steers how it assembles them.
A LangGraph writing assistant that interviews you before drafting and gates its output behind your approval — interviewer, writer, and voice critic as separate roles in one checkpointed graph. Play the human in a real recorded session in the interactive demo. The repo is private while a public release is prepared, so the demo is the tour — check back soon.
miseAn AI wardrobe assistant that outsources getting dressed: a vision model tags your closet once, then a daily email suggests weather-aware outfits and learns from thumbs-up/down feedback.