Client · 2021–2025 · Lakehouse
Two years of stalled migration. ETL from 15 hours to 20 minutes. Live four months later.
15h → 20m
ETL processing time
€400k/yr
Recurring licence cost eliminated
5,000
Users, zero disruptions at cutover
18%
FTE efficiency, holding-wide
Stepped in as data architect on a stalled migration. Redesigned the star model, phased cutover with rollback per layer. Zero disruptions for end users.
Migration map
Hover over a component to see what was going on, and what changed.
Old landscape
Oracle → Redshift lift-and-shift, stalled
Legacy source
Oracle
15-year-old ETL logic. Lift-and-shift to Redshift without redesign. The root of the migration crisis.
ETL layer
Redshift views (lift-and-shift)
Full loads over billions of records, no incremental logic. Every run: 15 hours.
Data Warehouse
Redshift
€400k/yr recurring licence costs. Weekly close unreliable for three months.
Reporting
MicroStrategy
20,000 dashboards in circulation. Nobody knew which were still actively used.
Phased cutover
Rollback per layer
New platform
Fast, scalable, explainable
Landing zone
S3
Validation against source at every layer. Errors caught early, not at the end user.
Data Warehouse
Snowflake
Incremental loading. From 15 hours to 20 minutes, structurally.
Transformations
dbt
Reusable framework blocks. One change automatically updates 12 downstream flows.
Reporting
Tableau
5,000 users. Zero disruptions at cutover.
What was replaced
What's there now
Click or hover for context per layer
Stack
Snowflake
dbt
Tableau
Airflow
OpenMetadata
Great Expectations
GitHub
Terraform
Approach
Lakehouse architecture
Data Mesh
Column-level lineage
Dynamic PII masking
TDD pipeline
Virtual DTAP · Zero Copy Cloning
CDC (Change Data Capture)
Matillion → dbt migration
Snowflake Cortex AI (POC)