The USDA provides a database of nutritional data, which is useful for food related software. It is structured as a zip of flat files and a PDF that describes the format.
This is a quick and dirty python script that converts that zip file to a sqlite
database. To avoid hard-coding schemas, I try to parse the schema from the pdftotext
output of the documentation.
To run this you will need the requests and sqlite3 python modules and the
pdftotext executable.
The resulting database uses text fields for NDB_NO, per the specification in the
PDF file. If you don't need an exact match for the upstream identifiers, which
have leading zeros, you could convert these fields to integers.