The realtime system is the heart and soul of Uber's logistic platform. It is responsible for fulfilling requests from riders while striving to maximize driver utilization and minimize rider cost. To make our realtime system efficient and intelligent, we need to extract deep and timely insights from our carefully curated data. We also have to make the insights easily accessible for both people and machines to consume in real time. This talk will discuss how stream processing is used in Uber's realtime system to solve a wide range problems, including but not limited to realtime aggregation and prediction on geospatial time series, data migration, monitoring and alerting, and extracting patterns from data streams. This talk will also present the architecture of the stream processing pipeline with a focus on how and why the architecture has evolved into its current form.