DuckDB is an analytical relational database management system which runs in-process, i.e., it is embedded within its host application that may be written in many popular languages (C/C++, Python, R, Java, etc.). DuckDB has full SQL support and can natively read/write formats such as CSV, Parquet, and JSON. It is built according to a modern system architecture, which allows running complex queries in parallel and spilling to disk for larger-than-memory workloads. This talk discusses DuckDB's key building blocks and demonstrates how it can handle hundreds of GBs of data on a laptop, or terabytes of data on a single server.