Snowpark and UDF/UDAF/UDTF

Snowpark brings a dataframe API to Snowflake so you can write transforms in Python, Scala, or Java that compile down to SQL and run inside the warehouse. When logic truly can’t be expressed as relational operations, you reach for user-defined code: UDFs (row-wise), UDAFs (aggregations), and UDTFs (table-returning). The guiding idea is simple: push work to the engine and keep data in Snowflake. Use Snowpark for orchestration-friendly pipelines; fall back to UDF-family only for the parts SQL can’t express cleanly.