Karel Coenye

As developers, we all love writing documentation… right? Error.... In reality, notebooks are where the best work happens—and where clarity often goes to die. Inconsistent structure, missing context, half‑explained decisions, and those “temporary” cells that somehow survive every refactor. Revisiting or sharing such notebooks weeks or months later becomes painful, and turning them into usable documentation is usually postponed… indefinitely. The intent of this session is to show a practical, repeatable approach to fixing this—by using Generative AI not as a code generator, but as developer assistant that supports both notebook quality and documentation across the software development lifecycle. The session demonstrates a workflow where existing analytical and engineering notebooks remain the single source of truth. Using a strong, carefully designed prompt, Generative AI applies consistent review pressure—much like an experienced colleague would during a thorough code review. From that reviewed notebook, the session then shows how documentation can be systematically derived, rather than written as a disconnected afterthought. The same source material is translated into multiple documentation layers, each tailored to a specific audience This is explicitly not a session about vibe coding or blind AI-driven generation. Human control remains central: developers own the code, the structure, and the final decisions. Generative AI is used to provide consistency, discipline, and scale in review and documentation—without introducing fragile or opaque workflows.

In this session, we explore the integration of Databricks Unity Catalog within the Flemish government. Let’s delve into how you can leverage Databricks Unity Catalog to effectively manage your data mesh and avoid the notorious data mess. I’ll explain how we achieve flexibility while maintaining a unified management infrastructure across our data and analytics workloads by dividing our agency’s data into governed data domains. These data domains, in turn, feed our data applications and AI models, providing valuable insights back into the core applications. Furthermore, we’ll take a deeper look at how we utilize workspaces as the primary analytical unit within our agency. These workspaces serve as gateways that map to our data mesh domains, allowing us to break down data silos even across different government entities.