I agree with everything in this article. If you're building from scratch, just store your data in Parquet and access it via Polars or DuckDB.

I built my own language for time-series analysis because of how much I hated q/kdb+, but Python has been the winner for a bunch of years now.