Glossary

What is Medallion Architecture?

Medallion architecture is a layered design pattern for a lakehouse or data lake. Raw data lands in a bronze layer as it arrived from source systems. A silver layer cleans, deduplicates, and conforms it into consistent tables. A gold layer shapes those tables for specific uses such as reporting, analytics, or machine learning features. Each layer reads from the one before it, so a team can trace a gold table back to the raw records behind it.

For example, a retailer might keep raw point-of-sale files in bronze, matched transactions in silver, and a daily sales-by-store table in gold. The pattern organizes refinement. It does not guarantee that data fits a given AI task. Steps that suit a dashboard can drop rare events, free text, or record-level detail that a model needs, so teams preparing data for AI check what each layer keeps and removes for that specific use.

Frequently asked questions

Is medallion architecture ETL or ELT?

It can use either. Most implementations load raw data into the bronze layer first and transform it inside the platform, which follows an ELT pattern. The layers describe stages of refinement, not a specific tool.

Why use medallion architecture?

It separates raw, cleaned, and business-ready data. That makes pipelines easier to debug, lets teams reprocess from raw data when logic changes, and gives each consumer a clear layer to read from.

Is gold-layer data ready for AI?

Not automatically. Gold tables are usually shaped for reporting. An AI task may need detail or rare cases that earlier steps filtered out, so readiness depends on the task the data will serve.