#graph

Mar 26, 2025

What We Learned Building a Real-World Fabric Data Agent

The post is a detailed, field‑tested account of building a real‑world Fabric Data Agent on top of a complex industrial supply‑chain dataset, highlighting what works, what breaks, and what requires deliberate engineering. It walks through six major lessons: why agent instructions—not data sources—determine answer quality; how Semantic Models and Ontologies complement each other; the pitfalls of asynchronous graph rebuilds; the importance of using binding property names in GQL; common GQL query‑generation failures; and the hidden requirement to use dbo. prefixes in SQL. Despite the rough edges of a preview product, the author argues that Fabric IQ’s architecture—unifying SQL, DAX, and GQL behind a governed semantic layer—is powerful and worth understanding deeply for anyone building enterprise‑grade AI analytics.

Ontology with Data Agent

This project provides a complete, end‑to‑end walkthrough for building a Microsoft Fabric Ontology demo—either by installing a ready‑made healthcare example or by creating your own custom ontology, lakehouse tables, demo data, and graph queries. It shows how to use tools like VS Code, GitHub Copilot, Semantic Link Labs notebooks, and the Fabric Ontology Playground to generate RDF models, orchestrate table creation, build a graph database, and construct multi‑hop GQL queries that highlight the strengths of graph reasoning. It also guides you through creating a Fabric Data Agent powered by your ontology, including AI instructions and sample prompts, enabling a fully functional demo environment in under an hour.