Definition
ETL transforms data before loading into the target. ELT loads raw-ish data first, then transforms inside the warehouse with SQL. Both move and reshape data reliably.
Orchestrators (Airflow, Dagster, cloud schedulers) run these jobs on schedules or events.
In simple terms
ETL is washing and chopping vegetables before putting them in the fridge. ELT puts the groceries away first, then prep when you cook.
Where you see it
- Nightly sync from app Postgres to BigQuery.
- Cleaning crawl data into training parquet files.
How it works
1.Extract
Pull from APIs, DBs, or files.
2.Transform or load
Clean, join, and type — order depends on ETL vs ELT.
3.Deliver
Write curated tables or ML datasets.
Why it matters
- ETL/ELT is how raw events become usable training and analytics data.
Often confused
ETL is only for business intelligence.
ML feature pipelines and corpus builders are ETL too.