Zalando uses machine learning for applications ranging from demand forecasting to recommender systems. In this post, Krysztof Szafranek describes the ML platform that enables this development. He discusses how Zalando data scientists use JupyterHub to prototype models when working with small datasets, and Spark when large scale data processing is necessary. To help users specify the AWS resources required for ML pipelines, Zalando created zflow – a DSL for concisely describing pipelines as a series of stages (zflow generates a CloudFormation template that creates a Step Functions workflow). Zalando tracks the execution of zflow pipelines through a developer portal created with Backstage.