While doing deep learning experiments, I noticed that it is very hard to write reusable code for training models. There are packages that help track metrics, logs, and checkpoints, but they often create more problems than they solve. As a result, training pipelines become bloated with infrastructure code that obscures the actual business logic.

That’s why I created this package, to build extensible training systems using domain driven design principles and replace ugly training scripts with clean and fully featured training services, with type annotations and modern python syntax.

Hope you find it useful.