Skip to content

lab

FinSight

A governed semantic model with its accuracy published and its scope visible, applied to finance reporting

What it does

A controller asks which cost centres moved more than 10% last month, and why. And gets an answer grounded in a governed finance semantic model, read-only, with the accuracy published and the scope visible.

Microsoft's Fabric Data Agent consumed by a Copilot Studio agent is their newest governed-data pattern and there is almost no complete public finance example of it. The agent itself does not exist yet: the spec gates the project on verifying that the paid Microsoft surfaces fit a personal budget, and that verification is a human's portal session. So this is deliberately the buildable half. The star schema, the 13 DAX measures, a 30-question accuracy suite with SQL-computed ground truth, and a governance transparency page, built so the portal session is short and scripted.

Walkthrough on request

Stack

  • Microsoft Fabric
  • Copilot Studio
  • DuckDB
  • DAX
  • Python

All data in this project is synthetic.