Popularized by Databricks, the Medallion Architecture is the default design pattern for modern Data Lakehouses. It organizes data into three distinct layers, each representing a higher level of quality and business value.
Bronze Layer (Raw)
The Bronze layer is the landing zone. Data here is ingested directly from the source systems (APIs, databases, Kafka streams, IoT sensors) in its raw state.
- Format: Usually Parquet or Delta, but keeps the original JSON/CSV schema.
- Rule: Never mutate the data. If a row is bad, keep it. You want a perfect historical archive of what the source system sent so you can always replay the history if a bug is found downstream.
Silver Layer (Cleaned & Conformed)
The Silver layer is the enterprise repository. This is where data is cleaned, filtered, and standardized.
- Transformations: Deduplication, casting strings to dates, handling nulls, and standardizing column names (e.g., changing
cust_idandCustomerIDtocustomer_id). - Usage: This layer is perfect for Data Scientists who want to train ML models on clean data, or for self-serve analytics where users know how to join tables themselves.
Gold Layer (Curated for Business)
The Gold layer is the presentation layer. The data is highly refined and aggregated for specific business use cases.
- Transformations: Complex joins across multiple Silver tables, aggregations (e.g., daily sales per region), and applying business rules.
- Usage: This data is directly queried by PowerBI or Tableau dashboards. End users (like Marketing or Finance) consume Gold data because it is ready for reporting and requires no complex SQL to understand.