Building NFL Analytics with Fabric, Power BI, and AI Agents

In this workshop, we build an end-to-end NFL analytics reference architecture in Microsoft Fabric using NFLVerse data. You’ll see how to download NFLVerse data locally with Python, validate the acquisition, create a Fabric workspace and Lakehouse, load raw files into Bronze, transform the data into curated Gold star schemas, validate the Gold model, deploy a Power BI semantic model, and prepare the solution for AI-powered analytics. The workshop also compares two Fabric Data Agent patterns: a Lakehouse Data Agent using SQL over Gold tables and a Semantic Model Data Agent using DAX over a governed Power BI semantic model. We’ll finish by testing the agents in Microsoft 365 Copilot, running evaluations, and walking through the agent development loop for improving natural language analytics quality. This project is intended as an educational reference architecture for data modeling, semantic modeling, Fabric Data Agents, and AI evaluation workflows. The NFLVerse data is used for demonstration purposes and should not be treated as production-audited sports data. GitHub repo: https://github.com/robkerr/nflverse-fabric-reference-architecture Chapters: 0:00 Introduction 1:42 Solution Architecture 3:59 Download Data 4:37 Run Local Tests 5:01 Create Workspace 5:36 Create Lakehouse 6:14 Upload Notebooks 6:53 Upload Raw Data 8:24 Raw Data to Bronze 9:32 Bronze to Gold 10:28 Validate Gold 11:26 Deploy Semantic Model 17:31 Prep for AI 19:49 Semantic Model Agent 23:47 Data Lake Agent 26:32 M365 Copilot Test 29:53 Agent Evaluations 32:05 Agent Dev Loop Topics covered: Microsoft Fabric, Lakehouse, Power BI Semantic Models, TMDL, DAX, Delta tables, Bronze/Silver/Gold architecture, NFLVerse, nflreadpy, Fabric Data Agents, NL2SQL, NL2DAX, Microsoft 365 Copilot, agent evaluation, and AI-ready data modeling.