Definition
A data lake keeps raw and lightly processed data (often in object storage like S3) so many teams can explore and transform it later. Schema is often applied on read.
Without governance, lakes become "data swamps" — hard to trust or find.
In simple terms
A warehouse stores labeled boxes on shelves. A lake is a huge yard where raw materials land first — flexible, but you must organize or you drown in piles.
Where you see it
- Storing web crawl dumps and audio before building NLP corpora.
- Clickstream and log archives for later analytics.
How it works
1.Ingest raw objects
Land files in cheap scalable storage.
2.Catalog
Register datasets, owners, and schemas.
3.Transform downstream
ETL/ELT into warehouses or training sets.
Why it matters
- Lakes are where modern AI training data often lives before curation.
Often confused
A data lake replaces the need for modeling.
You still need cleaning, labeling, and clear contracts for production ML.