A reusable sklearn preprocessing pipeline for the same UCI Adult Income data, built to run locally and as a SageMaker Processing job without modification. This is the project that makes the next ones honest — if features are computed differently in training than at serving time, every model after this is quietly wrong in production.
A single sklearn pipeline that works unchanged as SageMaker Processing is a direct, practical answer to 'how do you avoid train/serve skew' — a question that comes up in nearly every ML systems interview.