CartoDB is growing to be one of the biggest mapping platform for the masses, being powered by a fully open-source stack, with PostgreSQL, PostGIS, Mapnik and Leaflet at its core. Our aim is to democratize map and geographical data visualization, making it easy for non-GIS people to create simple maps using the CartoDB Editor, but still keeping all the power and flexibility of the underlying components available to advanced users, with a variety of building blocks ranging from the frontend with CartoDB.js and Torque to the backend with the Map, SQL and Import API, parts of what we call the CartoDB Platform. Serving dozens of millions of map tiles daily has its own set of problems, but when they are being created by hundreds of thousands of users (which have their own database and can alter everything from styling, to the data sources and the SQL queries applied) everything turns out to be a big source of challenges, both development and operationally speaking. This talk will go through our general architecture, some of the decisions we’ve had to take, the things we’ve learned and the problems we’ve had to tackle through the way of getting CartoDB to scale at our level of growth, and how we're giving back to the community what we've discovered though the process.