Resources
Everything here is free and open source. The labs and accelerators use synthetic data, so you can run them in your own Fabric tenant.
Data quality for the age of agents: the same batch of data can be safe for an LLM agent and harmful for an ML model.
Python package
Profiles how sensitive each data consumer (an ML model, an LLM agent) is to each kind of data defect, then checks every new batch per consumer without labels, with a pass, warn or fail verdict.
For: Data and ML engineers running models and agents on the same tables
Paper code
The corruption library, paired ML and agent evaluation harness, and sweep results behind the paper on why the defects that break ML pipelines differ from those that break LLM agents.
For: Researchers and practitioners who want to reproduce or extend the results
Small, practical things you can use this week.
Interactive assessment
Decide whether your data is ready for one specific AI use case. 18 questions across six layers (Sourced, Traceable, Retrievable, Aligned, Timely, Audited), six separate scores, open critical gaps and a 30/60/90-day plan. Runs entirely in the browser.
For: Data leaders scoping an AI assistant or RAG chatbot
Fabric notebook
Runs the Best Practice Analyzer on a semantic model, turns the findings into an editable fix plan, shows a dry run, then applies only the fixes you approved. No Tabular Editor needed.
For: Power BI and Fabric developers maintaining semantic models
End-to-end builds on Microsoft Fabric with synthetic data, so you can run them in your own tenant.
Workshop
Nine modules from three mirrored sources through a Bronze, Silver and Gold medallion, AI enrichment with Fabric AI functions, a Direct Lake report, a Fabric ontology and an ontology-grounded data agent.
For: Teams learning Fabric end to end, from ingestion to data agents
Workshop
A smart-city scenario in 12 modules across five goals: ingest, transform, build a report, apply AI and set up source control, with simulators for streaming data.
For: Software vendors and teams building analytics products on Fabric
Accelerator
A governed multi-agent copilot: three Fabric data agents answer as the signed-in advisor, a Foundry router decides who to ask, and a policy knowledge base adds cited rules. Includes the data generator, Direct Lake model, ontology and a Streamlit app.
For: Architects designing agents over governed enterprise data
Accelerator
Predicts each student's course outcome from LMS activity and explains every prediction, using time-aware weekly features with no look-ahead leakage, a calibrated model published with MLflow, and a model card.
For: Teams building explainable ML on Fabric
Guided lessons, free and self-paced.
New tools and labs land in the monthly brief
Key insights on Microsoft Fabric and AI, the month's in-depth articles, and the platform updates that matter. One email a month on LinkedIn.
Disclaimer
This is independent educational content and reflects my personal views, not those of Microsoft or my employer. Features and pricing may change, so verify important details in official documentation.