Pick an architecture to build
Build one, ask it questions, then try the other and compare the answers. The difference between them is the point of this lab.
Semantic Model Only
Upload a CSV, instantly create a semantic model, and connect it to a data agent. Fastest path to chat with your data. The agent reasons over tables and joins, the way Power BI does.
Ontology + Semantic Model
Define business entities and relationships, bind them to your data, and publish a knowledge graph the agent traverses. The full Fabric IQ loop, with the highest answer accuracy.
Ask an LLM about raw tables and it guesses.
Ground the same model in a knowledge graph built from your data and it answers with facts it can show you. This lab walks you through building that graph from plain CSV files, then lets you interrogate it.
▌
The left reply is a typical ungrounded answer. The right one is what this lab's data agent returns over the bundled demo data (15 orders, 8 customers, 6 cities). You can rebuild this exact exchange in the Data Agent step.
Choose Your Path
The concept behind this step
There are two ways to ground a Data Agent in your data. This lab lets you build both so you can compare answer accuracy.
- Semantic Model Only: Upload CSVs, define table-to-table relationships (like Power BI), and chat. Fast, but the agent only knows the raw tables and joins.
- Ontology + Semantic Model: Build a business ontology and knowledge graph first, then layer the model on top. More work, richer business context, fewer hallucinations.
Try the same questions on both and see which answers better.