AI_PARSE_DOCUMENT() — get your PDF invoices into the database

The new ai_parse_document() function enables us to detect text in various document types, including PDFs and images. I decided to test it by converting PDFs with invoices to a database-friendly format. To read PDF documents, we need to read files in binary format and then use the ai_parse_document function, which converts them into a string value and stores it in a table.

If you are not yet a member of Medium, you can access the extended version with video on the SunnyData blog for free.

Here is the basic syntax to achieve it:

WITH all_files AS (
  SELECT
    path,
    content
  FROM
    READ_FILES("/Volumes/hub/default/invoices", format => 'binaryFile')
)
-- Parse the files using ai_parse document
SELECT
  path,
  ai_parse_document(content) as parsed
FROM
  all_files;

output of the above SQL

The easiest way to create the above syntax is to go to “Agents” and set the PDF reader from our volume. I uploaded 3 example PDFs to the volume. “Use PDFs in Agent Bricks” will create a ready SQL script to read, and the script will be much longer in the above example, as it will also generate “cleaned” source data, which looks like a markdown invoice.