5X on count distinct is definitely good. The base dataset is 10M rows it looks like. The fact that this tool exists and has 3.5K stars on github is an indictment of the difficulty of getting started with hosted dbs, while also a wink to the current speed of broadband and power of memory in desktop machines now.

DuckDB is a sweet API for most programming languages and allows you to write SQL queries against parquet files, exploiting columnar storage (SQLite uses row storage).