One of the key lessons that we have learned along the way is that it's important to think about machine learning serving, about ML serving, not just about ML training. When you say machine learning to most people, they think about the complex pipeline on the left of this diagram. It's certainly where you, as a data engineer or a data scientist, will spend a lot of your time. However, the key reason you're doing machine learning is on the right-hand side of this diagram. You want to serve out those predictions to decision makers using notebooks, dashboards, applications, reports. Operationalizing a machine learning model, by which we mean, picking a model that's trained and getting to the point where we can serve out these predictions. Operationalizing a machine learning model is hard, and many projects fail to make it to this prediction stage. One of the lessons that we at Google learned was that in order to reduce our chance of failure, we needed to make sure that we could process batch data and streaming data the same way. Cloud Dataflow in this diagram, its open source is Apache Beam. Cloud Dataflow helps us treat batch and stream the same way. So Cloud Dataflow is just one example of how, on Google Cloud, you get to take advantage of our experience, Google's experience, in building machine learning infrastructure. If you haven't taken our data engineering specialization on Coursera, I strongly encourage you to take it. But in this track, we'll cover the key pieces as we go along. Fortunately, for those of you data scientists out there, data engineering is not that hard to learn. On GCP, the key services are all serverless and they're all managed infrastructure. So in this course, we will show you how to build batch and streaming data pipelines. By building your data pipelines on Google Cloud, you essentially get to take advantage of the scalability, reliability, and sheer engineering prowess that Google brings to running machine learning systems.