Tier-1 European Telecom Fabric Migration PoC
Teradata to Fabric. Rebuilt as Bronze, Silver and Gold.
Teradata-to-Microsoft Fabric modernization PoC using metadata-driven ingestion, Bronze/Silver/Gold processing, CDC/SCD Type 2, Delta tables, and Lakehouse/Warehouse architecture.
- Role
- Migration engineer
- Fabric ingestion, transformation, CDC/SCD2, validation
- Stack
- Microsoft Fabric
- Data Factory, Lakehouse, Warehouse, PySpark, Delta
- Scale
- 3 source systems
- Roughly 18-24 daily incremental source tables
- Outcome
- Reusable migration path
- Metadata-driven ingestion and reporting-ready serving layers
The brief. What it took, and what it changed.
Problem
Modernize legacy telecom data flows from Teradata/BTEQ and mixed source formats into Microsoft Fabric while preserving validation, retryability, and reporting readiness.
System
Fabric Data Factory pipelines and notebooks orchestrating source ingestion, landing, transformation, validation, Gold loading, and Warehouse serving across dual Lakehouse/single Warehouse architecture.
Impact
Supported 3 source systems and roughly 18-24 daily incremental tables; strongest scale and enterprise data-engineering credibility story.
How it works. In 4 stages.
Legacy extraction
Teradata, BTEQ, and mixed source files move into reusable Fabric ingestion paths.
Medallion processing
Bronze, Silver, and Gold layers preserve validation, retryability, and reporting readiness.
CDC and SCD2
Hash comparison, validity windows, active flags, and deleted tracking keep incremental history consistent.
Warehouse serving
Lakehouse and Warehouse layers prepare migrated data for downstream analytics consumption.
What shaped the build. And what it runs on.
Used metadata-driven ingestion for reusable migration flows.
Applied MD5 hash-based CDC, rec_start/rec_end validity windows, active/deleted tracking, schema correction, UTF-8 standardization, JSON flattening, and idempotent execution patterns.
Built with
- Microsoft Fabric
- Fabric Data Factory
- Lakehouse
- Warehouse
- OneLake
- PySpark
- Spark SQL
- SQL notebooks
- Delta tables
Proof. What's public, and what's on its way.
References
- Project linkPublic architecture or validation example to add after approval.Pending
Visual artifacts
- Architecture diagramdiagramTo approve
- Medallion flowdiagramTo approve
- Sanitized validation screenshotsscreenshotTo approve
Private project summaryMetric
Supported 3 telecom source systems in a Microsoft Fabric migration PoC.
- Source
- Resume extraction and project inventory
- Confidence
- High confidence
- Public note
- Client identity remains anonymized as a Tier-1 European telecom provider unless permission is confirmed.
Private project summaryMetric
Supported roughly 18-24 daily incremental source tables using metadata-driven ingestion and Delta processing.
- Source
- Resume extraction and project inventory
- Confidence
- High confidence
- Public note
- Keep source-system names and private pipeline details out of the public version.
Tradeoffs. And what stays private.
The tradeoff
Detailed source-system names should stay out of the public case study unless client permission is confirmed.
Kept anonymous
Keep client anonymized as a Tier-1 European telecom provider unless permission is confirmed.