Post

the sustainable angle is what sold me on smaller models before efficiency was fashionable. in applied ML most of my wins came from feature engineering and a right-sized model, not scale. the failure mode is reaching for GenAI when a linear model would ship faster.