The medallion architecture has become the go-to pattern for organizing data in a lakehouse, but ask five people how to implement it, and you'll get seven different answers. Where does the business logic go? What belongs in silver? Should bronze even have Delta tables? If you've ever felt confused by contradictory advice on how to layer your data platform, this session is for you.

In this session, I'll walk you through how I implement the medallion architecture in Microsoft Fabric and more importantly, why. This is not a "it depends" kind of talk. I have strong opinions on what each layer should and should not do, shaped by years of building production data platforms. I'll share those opinions openly, explain the reasoning behind them, and show you what this looks like in practice inside a Fabric lakehouse.

We'll start at the bronze layer, where raw data from APIs, databases, and files lands in the Files section of a lakehouse before being structured into Delta tables. Then we'll move to the silver layer — and this is where my approach breaks from the crowd. No business logic. No transformations. No aggregations. Silver is a faithful, versioned, schema-managed mirror of your source systems, enriched only with lineage and system fields to support time travel and schema evolution. Finally, we'll arrive at the gold layer, where PySpark notebooks (or stored procedures) transform silver data into dimensional models following the Kimball methodology: ready to power Power BI semantic models, AI data agents, and beyond.

Whether you're a Power BI professional stepping into data engineering, a data engineer exploring Fabric for the first time, or someone migrating from another platform, you'll walk away with a clear, practical, and opinionated blueprint for structuring your data lakehouse. A basic understanding of data warehousing concepts and a general awareness of what Microsoft Fabric is will help you get the most out of this session.

After attending, you will be able to: explain the purpose and responsibilities of each medallion layer; design a bronze layer that separates raw file ingestion from Delta table structuring; implement a silver layer focused on source fidelity, schema evolution, and time travel — without business logic; build a gold layer using notebooks and stored procedures to deliver Kimball-style dimensional models; and make deliberate architectural choices about where logic belongs in your data platform.