If you have a database that supports SQLMED, you can probably already query outward table data without loading/serialising between formats. You can even optionally generate an index on some systems.

MariaDB has the Connect Engine, Postgres has Foreign Data Wrappers (fdw), define an openrowset for mssql using an odbc or ole driver and config. DB2 and informix have federated options.

The author has missed a lot of detail about the breadth of options the FDW for csv provides. Just like a normal table you should define column names and ideally their types so you don’t have just ordinal/generates column names (if there is no header line), which the author does anyway for SQLite. The same problem of denormalisation exists, in which case, data scrubbing cannot be avoided; yet more serialisation eating precious cycles and memory.

My point is, read the manual for your database, it might surprise you as to what it can already do, and how much time it can save you.