Abstract

This script does the following:

• takes a folder full of badly-named PDFs,
• extracts text from the first few pages,
• sends that text to Anthropic's smallest/cheapest model to guess the title, author and date of first publication
• and renames files with the results if results are satisfactory

Methodology

I wrote the basic scripts with the help of Claude+Gemini, and then tweaked the system prompt to reflect my specific renaming conditions.

Results

I take a folder full of PDFs named in haphazard ways like (12345.pdf or 2023-report-OECD.pdf) and it gets renamed as (OECD - Annual Report for the Pacific (2021).pdf)