The problem

The warehouse had grown around a single use case. Adding a new domain meant reworking the whole stack.

Key decisions

  • Separate raw data from business semantics.
  • Incremental dbt transformations run on ECS, orchestrated through Glue.
  • Physical design tuned to workload, not one-size-fits-all.
  • Marts aligned to consumption domains with clear ownership.

Outcome

New domains can be added without recreating the stack, and the warehouse serves the wider company rather than one product.

RedshiftdbtMedallionData Marts

Architecture flow

  1. Source systems
  2. Ingestion
  3. Bronze: raw and traceable
  4. Silver: clean and standardized
  5. Gold: business models
  6. Data marts for domain consumption