Writing
Short notes on research projects, evaluation craft, and the unglamorous middle of getting machine learning systems to actually be useful.
Why working in IT infrastructure rewards the same habit ML engineers should be cultivating: a mental simulation of the system you're touching, updated faster than the docs.
A scaled-down medical scribe assistant: notes, design decisions, and what I had to throw away before the thing felt useful.
BLEU, F1, AUC, and confusion matrices say different things on a few thousand examples than on a million. A note on which numbers to trust, which to caveat, and which to ignore when you're working at student-project scale.
What unsupervised methods are still good for in 2026, and how Self-Organizing Maps quietly carried the climate-pattern work that the CNN got most of the credit for.