Presentations

Source Control in Power BI: a BI developer’s journey to CI/CD, Athens Fabric & Power BI Meetup, March 2025.

A talk arguing that Power BI development deserves to be treated as software engineering, and that Microsoft’s new project format finally makes it possible, opening a traditionally closed, proprietary Fabric environment up to version control, review, security, and even AI agents. I built and delivered it the way I build documentation: a structured, sourced case aimed at moving a community of practitioners toward a more open and accountable way of working.

What it covered:

  • The problem. BI developers are “software developers supercharged with business knowledge”, yet were stuck editing binary .pbix files, versioning by saving copies, and exporting to Excel, with no diffing, no review, no safe collaboration.

  • The grounding. I set the argument on the business-intelligence-systems literature (Davenport and Arnott; the finding that roughly 70% of BI projects fail to meet expectations without proper foundations) and closed by mapping the approach to the DeLone and McLean information-systems success model.

  • Microsoft’s “revolutionary trinity”. The PBIP structure with the TMDL text format (turning binary models into transparent, version-controllable, human- and agent-readable folders), Git integration with Power BI Service workspaces, and Copilot integration.

  • A metadata-only security model. Models reference data sources by name only, so no production data ever travels inside a file, which satisfies strict compliance requirements.

  • The real implementation at Maya: a shared “universal” core model, automated workspace and customer-folder creation via the Power BI API and GitHub Actions, a branch → test on a dummy workspace → pull request → merge loop, a standardized .gitignore, and Python automations (for example, rebinding report visuals straight from tmdl metadata).

  • An agent-experience outlook. Because source-controlled Power BI artifacts are granular, structured text (JSON plus tmdl), agents can parse and even generate reports without proprietary connectors.

Slides · meetup page