Good software engineering is crucial to building a successful small machine learning system beyond just the machine learning model, managing the change process, and deploying it effectively. Dealing with test set distributions that are significantly different from the training set distribution in machine learning is still a challenge for academia.