Every trip I make with my camper van gets logged: kilometers, destinations, costs. Nothing complicated, which made it a good test case for something bigger: could a full Power BI build, from a basic semantic model to a finished dashboard, be handed over to Claude and still come out right?

This session walks through that build: Power BI Desktop, VS Code, Claude Desktop, the MCP modeling server, and Microsoft's Skills for Fabric.
I'll show Claude asking clarifying questions before writing a measure, flagging errors in the dataset, and building a custom theme from a photo of my van.

I'll also show where it broke down: a KPI visual built the wrong way, and the fix that mattered most wasn't the fix itself; it was what came after. I corrected it once by hand, pointed it out in the code, and had Claude apply that fix everywhere else the problem showed up.

Attendees leave with a workflow to try themselves, and one habit: correct once, then teach the correction forward.