Field notes / 001
A working archive of what I’m figuring out.
I build production ML systems, read papers, and write down what stuck. This is the long-form margin where half-formed ideas go to get tested in public.
What I'm into right now
03 — THREADS
THREAD 01
Explainability that survives production
Concept-based explanations, drift monitoring, and what "interpretable" means once real users depend on the model.
THREAD 02
Trustworthy ML, the security way
Applying threat-model thinking from fraud prevention and cybersecurity to how ML systems get evaluated.
THREAD 03
Data infrastructure for ML teams
Lakehouse tables, fast loading paths, and deployment pipelines that ML teams own end-to-end.