Machine learning algorithms require fixed-size feature vectors. It is why modern deep learning architectures deliver sensational results on audio, vision, and language but fail on relational data. The only way out is feature engineering. Unfortunately, manual feature engineering requires a lot of SQL code and takes up to 90% of a project's time. In this talk, I want to introduce you to the domain of relational learning. Attendees will learn how to build end-to-end prediction pipelines on relational data using algorithms for automated feature engineering.