KNIME to Microsoft Fabric Migration, Accelerated with Claude AI
Hundreds of drag-and-drop KNIME workflows had to move to Microsoft Fabric. Rebuilding each one by hand, node by node, is slow and easy to get wrong, so I paired a hands-on migration with an AI-assisted accelerator built on Claude and the Fabric MCP server.
- KNIME workflow (.knwf)
- Claude Code
- PySpark notebook
- Human review
- Fabric MCP
- Fabric Lakehouse
- Led the migration of 250+ legacy KNIME workflows to Microsoft Fabric across 6–8 sources, including PostgreSQL and Amazon Redshift.
- Rebuilt ingestion, cleansing and business logic with Dataflows Gen2, PySpark notebooks and Fabric Pipelines in one unified Lakehouse.
- Validated every output source-to-target against the legacy KNIME results, reaching 99.99% data accuracy.
- Cut manual effort and processing time while keeping full business-logic parity.
- Used Claude Code to read each KNIME workflow, put its nodes in the right order and generate a ready-to-run PySpark notebook, with one labelled section per original KNIME node.
- Connected Claude to Microsoft Fabric through the Fabric MCP server, so reviewed logic could be pushed straight into the Fabric Lakehouse and Dataflows.
- Unsupported nodes become clearly marked to-dos instead of stopping the run; custom Java, R and Python snippet nodes were finished and reviewed by hand.
- Shifted the work from rewriting every node by hand to reviewing and refining generated code, checked side by side against the original workflow.
One KNIME workflow, rebuilt as 13 Dataflow Gen2 stages
Every stage hands off to the next through Bronze Lakehouse tables. Hover or tap a stage to trace exactly where its data comes from and where it goes.