When performing data transformations and cleaning, do you typically handle them in separate stages or within the same pipeline?

I usually separate extraction, transformation, and loading into distinct stages to keep the pipeline modular and maintainable. Extraction pulls raw data into a bronze layer (raw JSON, CSV, etc.). Transformation is performed in a silver layer using tools like dbt, Spark, or SQL scripts, where I apply cleaning rules, type casting, deduplication, and enrichment. Loading moves the cleaned data into a gold layer or a data warehouse for analytics. This separation allows me to run data quality checks after each stage, roll back changes easily, and reuse transformation logic across multiple downstream consumers. In smaller projects, I might combine extraction and transformation in a single job, but for production systems the staged approach scales better.