A machine learning pipeline is a sequence of stateful transformations. Understanding how to compose fit/transform steps, avoid leakage across pipeline boundaries, and apply transformations consistently between training and inference separates toy experiments from production systems.