Most programmers have used a database at some point in their career. And while there's a lot of talk about features and languages and the differences between SQL and NoSQL, few people know what actually happens inside those databases. In this talk, we will peel off some of the outer layers and take a deeper look into database internals and their storage engines, exploring the concepts in an intuitive way. The goal is to provide the audience with a better understanding of databases in general to help them make better choices for their own projects and to explain why the next shiny NoSQL database that arrives may in fact not be as unique as it may seem at first glance.

Over the years, analytical data systems have evolved from data warehouses to data lakes and, eventually, lakehouses. Along the way, they gained powerful features, but also increasing complexity. For many teams, this creates an awkward choice. You can keep things simple with Postgres, but miss out on features like snapshots, schema evolution, and time travel. Or you can adopt a modern lakehouse stack, only to discover that it comes with catalogs, metadata layers, and enough moving parts to feel like a platform project in its own right. But what if there was a middle ground? In this talk, we’ll take a look at DuckLake, a new format released in 2025 by the DuckDB team, that tries to simplify the lakehouse architecture by treating metadata as a database problem. The goal is to keep the useful parts of modern analytical systems while making them easier to understand, operate, and grow with from local development on a laptop all the way to large-scale cloud workloads. We’ll explore how DuckLake works, what trade-offs come with adopting this new approach, and whether it can really provide a simpler path between “just use Postgres” and running a full-blown lakehouse platform.