Data & Analytics · Open-source build
Analytics lakehouse with tested transformations
Raw data lands once, every transformation is versioned and tested, and every dashboard number can be traced back to its source rows.
What it is
Dagster orchestrates ingestion into object storage as Parquet. dbt models build staging, intermediate and mart layers with tests on uniqueness, nullability and referential integrity, and the semantic layer defines each metric exactly once. DuckDB serves interactive queries; the same models run on a warehouse without change.
What the build consists of
- 01Ingestion assets with freshness checks and lineage in Dagster
- 02dbt project with staging, intermediate and mart layers and tests on every model
- 03Metric definitions in the semantic layer, consumed by the dashboards
- 04Column-level lineage from dashboard tile to source table
- 05Data quality alerts routed to the owning team
- 06Cost and runtime per model tracked over time
What you see in a demo
Clicking a number on a dashboard and walking back through the models to the raw rows that produced it, then breaking a source and watching the test catch it before anyone sees a wrong chart.