Feature Engineering

The feature discussion is the single easiest place to lose an ML design interview. There are infinitely many plausible features, so candidates either dump thirty of them in a stream of consciousness or freeze because they do not know the domain. Both fail for the same reason: no structure. This page gives you the structure: a small taxonomy of signal sources, the modern encodings for each data shape, the freshness ladder, and the pitfalls interviewers use as traps.