If you're in the financial services industry or have an interest in predicting market movements with machine learning, you may be eager to learn how to move your trading signal and forecast generation code into the cloud. You can easily scale up your computational loads, distribute data processing pipelines to run in parallel on multiple machines, speed up the time required to run complex analytics, eliminate the need for management of data storage, and ultimately eliminate the need for multiple data centers. In this post, we'll show how to build a data processing pipeline that starts with a market data feed as the input and uses machine learning to generate real-time forecasts as the output, with all application components running natively in Google Cloud. In the sections below, you'll learn how to build a complete end-to-end application that subscribes to the Thomson Reuters FX (foreign exchange) data feed published on a Cloud Pub/Sub topic, incrementally trains a TensorFlow neural network model, generates real-time forecasts of FX rates, and saves the forecasts into BigQuery for subsequent analysis. First, we'll focus on Cloud Pub/Sub as the connector used to link multiple application components.