In this post, we explore how to build practical and scalable recommendation systems using Neo4j and its Graph Data Science library. We start with a fundamental example of news recommendation on a dataset containing 17.5 million click events and around 750K users, leveraging Neo4j's graph data model to quickly predict similar news based on user preferences. The post defines the basics of Recommender Systems, explains how they work, and provides an overview of the Microsoft MIND dataset used in this example. We also cover a technique called Collaborative Filtering (CF) which is applied using Cypher query language and scaling with the Graph Data Science library, leveraging node embeddings and an ML technique called K-Nearest Neighbor (KNN). The post concludes by discussing next steps and providing resources for further exploration of graph-based recommenders.