Duckdb and polars are great

For $work, we use SQL Server for data pipelines. It's quite capable, but doesn't work well when your workload requires more than a single node of compute.

duckdb and polars offer great interop with files (particularly parquet) and good default implementations of most important features.

It's nice that you don't have to host a persistent server; just load up the binary and start querying data.