Skip to content

Case study

Standing up a governed Fabric platform, project code PROJ

Written generically and without identifying detail. The interest is in the method, not the organisation.

Context

Placeholder body copy to be completed.

Starting position

Placeholder body copy to be completed.

Constraints

Placeholder body copy to be completed.

Approach

Placeholder body copy to be completed.

What changed

Placeholder body copy to be completed.

Handover

Placeholder body copy to be completed.

Final phase: the data agent

The natural-language data agent was the last phase of the build, deliberately. It was only grounded once the Gold layer was conformed, the measures were documented with named owners, row-level security was enforced at the model and the verification tests were passing on the layer the agent would read.

The agent answers from that governed semantic model rather than from raw tables, and it is reached inside the reporting experience the business already opens rather than as a separate chat tool. The sequence was the point: govern the model, then ground the agent, then surface it where people work.

A Fabric app for the workflow the report canvas could not express

Part of the same engagement was a custom application built on the Rayfin SDK, running inside Fabric in a workspace alongside the data. It reads from the governed semantic model rather than from a separate copy, so the definitions and the access rules already agreed apply to the application as they do to reporting.

The application was authored, versioned and promoted through the same pipeline as every other artifact in project PROJ, which meant it inherited the branch-per-environment rule and the verification step rather than being deployed by hand.

Want the same starting point?

Book an architecture review