Core message of the talk: developers are tasked with solving problems, not with implementing a particular canonical solution. Sometimes proper machine learning is the right tool and sometimes it isn't — Nebulab's coffee-box problem was small enough to reason about and benefit from a hand-crafted, auditable scoring engine that looked like a neural network but wasn't one. Using a framework-less scoring approach let them iterate quickly and explain decisions to stakeholders, at the cost of hand-tuning coefficients.