Checklist — AI semantic model setup

Power BI / Fabric

Use this checklist to set up an environment where you can generate and improve semantic models with AI. Tick off the steps as you go.

1. Set up a Power BI Project (PBIP)

Create a new Power BI Project (PBIP)
Open the project locally on your machine
Make sure your semantic model is visible as files (TMDL)

More info: Power BI Projects overview

2. Basic knowledge of TMDL

Understand how tables, relationships and measures are structured in TMDL
Review how metadata and descriptions are stored
Open existing TMDL files and read the structure

More info: TMDL documentation

3. Visual Studio Code environment

Install Visual Studio Code
Open your PBIP project in VS Code
Verify you can edit and save files

Download: code.visualstudio.com

4. GitHub and repository

Create a GitHub repository
Connect your local project to the repository
Perform an initial commit and push

More info: GitHub quickstart

5. Activate GitHub Copilot

Install GitHub Copilot in Visual Studio Code
Sign in with your GitHub account
Test whether Copilot can generate code in your project

More info: GitHub Copilot documentation

6. Install an MCP server (Power BI / Fabric)

Install an MCP server that works with Power BI or Fabric
Configure the connection to your model
Test whether AI can perform actions on your model

More info: MCP servers overview

7. Define instructions and skills

Create an initial instruction file (skills)
Define which business terms you use
Define model rules: star schema, naming, measure structure, synonyms and descriptions

Tip: start small and expand iteratively.

8. Generate your first model

Have AI generate tables and relationships
Check whether the model is logically structured
Review measures and definitions
Verify naming and structure

9. Iterate and improve

Adjust your instructions based on output
Regenerate the model
Compare results and improve consistency

10. Ready for use

Model usable for reporting
Synonyms and descriptions in place
Structure suitable for Data Agents

This process is iterative. The first version will rarely be perfect. The power lies in improving your instructions and repeatedly regenerating the model.

Receive this checklist as a PDF?

Click below to receive the checklist by email as a PDF.

Your email address will be used to send the checklist and may be used for relevant follow-up. Your data will not be shared with third parties and is carefully managed by Dogoda. Read more about privacy.

Back to the note on AI and data models, or view the overview page of all Gebruikersdagen notes.