Direct Lake changes the maths
Direct Lake keeps reports reading straight from Delta parquet — large models stay fresh without a full import.
All projects
Data Engineering Practice
Microsoft Fabric Data Engineering Practice
An end-to-end lakehouse pipeline — and a Fabric learning ground.
- Microsoft Fabric
- Dataflow Gen2
- Lakehouse
- Delta Lake
- Fabric Pipelines
- Power BI
Overview
A hands-on, end-to-end project on Microsoft Fabric: a metadata-driven ingestion pipeline that lands source data in a lakehouse, refines it through medallion layers and serves it to Power BI through a Direct Lake semantic model.
The project is also a deliberate learning vehicle — building the Fabric skills (lakehouse architecture, Dataflow Gen2, pipelines, Delta Lake) that increasingly sit underneath Power BI.
The business problem
Classic Power BI projects often stop at Power Query and import mode, which limits how far the refresh architecture can scale.
The challenge here was to explore how the Fabric lakehouse changes that pattern — ingesting, transforming and serving data with lakehouse tooling while keeping the reporting layer firmly in Power BI.
Fabric pipeline
Metadata-driven
Dataflow Gen2
Ingest
Lakehouse
OneLake
Semantic model
Direct Lake
BronzeSilverGold
Objectives
- Build a metadata-driven ingestion pipeline instead of one-off copies
- Implement medallion architecture with bronze, silver and gold layers
- Serve Power BI through a Direct Lake semantic model
- Document the learnings for future Power BI + Fabric projects
Data architecture
- 01
Ingestion
Metadata-driven Dataflow Gen2 ingestion lands source data in OneLake.
- 02
Medallion lakehouse
Bronze, silver and gold lakehouse layers refine raw data into reporting-ready Delta tables.
- 03
Orchestration
Fabric pipelines orchestrate refresh and dependency ordering across layers.
- 04
Direct Lake model
A Direct Lake semantic model reads Delta parquet directly — no separate import step.
- 05
Power BI reporting
The model is consumed exactly like any Power BI semantic model — reports, DAX and row-level security.
Dashboards
Fabric pipeline
Metadata-driven
Dataflow Gen2
Ingest
Lakehouse
OneLake
Semantic model
Direct Lake
BronzeSilverGold
Fabric pipeline
Metadata-driven
Dataflow Gen2
Ingest
Lakehouse
OneLake
Semantic model
Direct Lake
BronzeSilverGold
Key insights
Metadata-driven scales
A metadata-driven ingestion pattern means adding a new source is configuration, not a rebuild.
Layering pays off
Medallion discipline (bronze, silver, gold) pays for itself the first time something upstream changes.
Outcomes
End-to-end pipeline
A working end-to-end Fabric pipeline, built and documented as a personal learning ground.
Direct Lake learned
First-hand experience with Direct Lake semantics feeding Power BI.
Reusable learnings
Documented patterns ready to apply on future Power BI + Fabric engagements.
See the report in action
Live report coming soon — this practice project will link directly to the published report.