Writing from the work.
Engineering decisions, research problems, and the convictions behind them — written from the projects themselves, including the parts that went wrong.
Spatial cross-validation when the data won’t spread out
Random k-fold overstates performance on clustered ecological data. The obvious alternatives broke on my dataset too — here is the fold design QueleaGuard ended up with, and why.
Read the postThe API chose my sampling design
How an undocumented sort order in a public biodiversity API put 97% of my background points in January — and what I changed so that kind of bias gets caught before a model ever sees it.
Read the postWhy I started Sauti Labs
It began with a speech model that didn’t understand the people I was building for. What I found when I went looking for alternatives, and why I think the distance between African-language research and usable products is worth building a company around.
Read the postAI as a detector, not a decision-maker
Designing Scribe AI’s live path: offline speech recognition, rule-based scripture detection, and a human approval step that keeps a wrong guess off a church screen.
Read the postCoverage is not capability: what Whisper taught me about African speech
A language appearing in a model’s supported list says little about whether the model will work for your speakers. Why Whisper struggled with the speech I needed, and what I now check before building on any speech model.
Read the postModelling the landlord who isn’t there
Most rental software assumes both sides of a tenancy are on the platform. That assumption fails for many tenants — so Rentra had to represent the relationship before the landlord arrives.
Read the post