The main obstacle to building good data communication products is that it’s difficult to predict how complex they will need to be in order to effectively transmit the value and limitations of your analysis. While you need to build the initial prototype very quickly, those prototypes often grow in scope to become large, mission-critical applications. It’s hard to find an application framework that allows your app to grow smoothly as its requirements increase.

Shiny for Python is ideal for building these applications because it does two main things. First, it uses a reactive computation graph to minimally re-render components. This means that you can build large applications which nevertheless run quickly and scale well. Second, Shiny infers this graph from your application, which means that you don’t usually need to manually cache data, manage state, or write callback functions. While this inference can seem like magic, it’s based on a simple and intuitive algorithm that lets your apps scale from prototype to product. This talk goes through the details of that algorithm to show why it’s the best way to build modern data science applications.