# Medallion data warehouse on an open lakehouse

> A medallion data warehouse on DataNature lands raw data in Bronze, cleans and standardizes it in Silver, and publishes certified dimensional and fact tables in Gold, as open Delta Lake or Apache Iceberg tables. Trino serves fast SQL, the semantic layer defines KPIs once, and Atlas tracks lineage from report to source.

Source: https://datanature.ai/solutions/medallion-data-warehouse
Publisher: Dlytica Inc. (DataNature is a Dlytica company)
Updated: 2026-11-05

Bronze, Silver and Gold on Delta or Iceberg tables, with no separate warehouse

## Components

- **Connect**: Batch, incremental and full loads from 200+ sources into Bronze
- **Data Engineering workbench**: dbt and Spark transformations from Bronze to Gold
- **Lakehouse**: Delta Lake and Apache Iceberg tables with ACID, time travel and schema evolution
- **SQL warehouse**: Massively parallel SQL with Trino federation
- **Semantic Layer**: Certified views and KPIs for BI, Voir and apps
- **Atlas**: Catalog, RBAC, PII masking, lineage and audit

## Steps

1. **Land in Bronze**: Connect loads sources as raw, immutable, timestamped and catalogued copies. Bronze is never deleted, so data can always be reprocessed.
2. **Clean in Silver**: dbt models deduplicate, handle nulls, check referential integrity and apply business rules.
3. **Publish Gold**: Dimensional and fact tables, certified and version-controlled, optimized for analytics.
4. **Define KPIs once**: The semantic layer exposes certified views and KPIs to BI, Voir and third-party tools.
5. **Govern and audit**: Every dbt run writes lineage to Atlas. Time travel makes reports reproducible.

## Documentation

- [Data engineering](https://docs.datanature.ai/docs/developers/data-management/data-engineering)
- [Data warehouse](https://docs.datanature.ai/docs/developers/data-management/data-warehouse)

## Why build a warehouse on a lakehouse?

Because you get one copy of data. The warehouse runs directly on open Delta or Iceberg tables, so there is no separate warehouse to load, no data movement, and no lock-in.

## What goes in each layer?

- **Bronze:** raw, immutable copies of every source, timestamped and partitioned.
- **Silver:** cleansed, deduplicated, standardized and validated against business rules.
- **Gold:** business-ready dimensional and fact tables, certified and versioned.

## How do BI tools and agents use it?

Through the semantic layer. Power BI, Tableau, DataNature BI and Voir all query certified views and KPIs, so every number matches.

## How long does a migration take?

It depends on the number of sources and reports. Dlytica Consulting scopes it in a short assessment and runs old and new in parallel, reconciling with Atlas lineage before cutover.

## Common questions

### What is the medallion architecture?

Bronze holds raw data, Silver cleaned data and Gold business-ready tables. See [medallion architecture explained](/blog/medallion-architecture-explained).

### Do we still need a separate data warehouse?

No. The SQL warehouse runs directly on open lakehouse tables, so there is one copy of data.

### Is a lakehouse overkill for small data?

Sometimes. See [the honest guide for small teams](/blog/is-a-lakehouse-overkill-for-small-data).

## Implementation details

1. **Land sources in Bronze** with Connect: Oracle, SQL Server, MySQL and PostgreSQL over JDBC into `lakehouse.bronze_<source>.<table>`. Use full loads for reference tables, Full → Incremental for large transactional tables, and a bookmark column that changes on every update.
2. **Partition only large tables**, by month for long histories; set partitioning before deploy.
3. **Model Silver in dbt** on the Spark target: de-duplicate to the latest row per key, cast types, and add `not_null` and `unique` tests.
4. **Model Gold marts per domain** (finance, customer, risk) as facts and dimensions, with `relationships` tests and incremental materialization for large facts.
5. **Define certified KPIs** as semantic-layer cubes over Gold, and materialize the busiest ones in DN Warehouse.
6. **Schedule one Ingest + Transform job** per domain: pipeline → dbt Silver → dbt Gold → Refresh cubes, in your time zone.
7. **Apply governance** in Atlas: groups mirroring the organization, PII masks in Bronze and Silver too, row filters on Gold.
8. **Reconcile** Gold totals against the legacy warehouse per period before switching reports over.
