Definitely checking this out today! I use postgres for ~30 GB of machine learning data (object detection) and have a couple workflows which go through the Postgres->Parquet->DuckDB processing route.

A couple questions, if you have time:

1. How do you guys handle multi-dimensional arrays? I've had issues with a few postgres-facing interfaces (libraries or middleware) where they believe everything is a 1D array!

2. I saw you are using pg_duckdb/duckdb under the hood. I've had issues calling plain-SQL functions defined on the postgres server, when duckdb is involved. Does BemiDB support them?

Thanks for sharing, and good luck with it!