#ai

Jul 23, 2026

E-commerce ontology and data agent in Microsoft Fabric IQ

Your data can be clean, complete, and connected, and still fail to answer a simple business question. Microsoft Fabric IQ solves this with something called an ontology. I built one end to end, connected a data agent to it, and wrote down every step. If you have never touched an ontology before, this guide is for you. Without an ontology, an AI agent can only guess at your data.

Sep 10, 2026

Why My Fabric Data Agent Returned the Wrong Number : Diagnosing a Silent Grounding Contamination Bug in Azure AI Foundry

This article documents a production debugging session where an AI agent connected to Fabric Data Agent returned an incorrect number due to a File Search knowledge file contaminating the data grounding. It explains how to diagnose this using Azure AI Foundry traces and the architectural fix (moving static reference data into the system prompt and enforcing strict tool-gating) that resolved both correctness and response latency.

Sep 14, 2026

Why Your Foundry Agent Works in the Playground But Fails for Every Other User The Identity Passthrough Problem Nobody Warns You About

This article documents why an Azure AI Foundry agent connected to a Microsoft Fabric Data Agent works perfectly for the builder but fails for every other user with tool_user_error or timeout errors. It explains the On-Behalf-Of (OBO) identity passthrough mechanism, maps out the four separate permission layers required across two different portals, and shows how a single Microsoft Entra security group fixes the problem permanently.

Agentic App with Fabric (incl Data Agents)

Agentic Banking App is an interactive web application designed to simulate a modern banking dashboard. Its primary purpose is to serve as an educational tool, demonstrating: How SQL-based databases are leveraged across different types of workloads: OLTP, OLAP, AI workloads. How agile AI-driven analysis and insight discovery can be enabled via prescriptive data models in Fabric. How easy it is possible to integrate other Fabric workloads (e.g., Report, Data Agent, Notebook) leveraging that data model. Through a hands-on interface, users can see the practical difference between writing a new transaction to the database, running complex analytical queries on historical data, and using natural language to ask an agent to query the database for them.

Apr 18, 2026

Data Agent to Data Memo

Experimented with a pipeline to explore the data autonomously, find insights, find second order signals, create a cohesive narrative that's grounded in the data & context, added an optimization layer using DSPy (Community) to tune it and converted to speech. No external APIs, services, 100% in hashtag#MicrosoftFabric . 🔊Listen, let me know. The novelty here is I used the recently published LLM-as-Verifier approach as a metric in GEPA to tune the narrative. (https://lnkd.in/g2Huycw9)