Paperless AI needs to support rescanning / OCR'ing documents via LLM

[anaisbetts]

Is your feature request related to a problem? Please describe.

Every feature in Paperless AI is contingent on the text of the document being correct - tagging, AI Chat, everything. Unfortunately, Tesseract OCR is actually Pretty Bad, and for many documents is just out there generating garbage. This means that everything built on top is functionally unusable - ask "what is my account number for $SERVICE" and it gives you literally the wrong number, because it was scanned poorly.

Describe the solution you'd like
Co-opt the features of Paperless-GPT only better - as part of the processing, also re-OCR the document using modern solutions and save that as the document text, then do tagging based on that.

Describe alternatives you've considered
Fixing paperless-gpt but that is not a very great project overall.

[beneiltis]

This! 💯 👍

[krizzpi1337]

This

[clusterzx]

This issue has been marked as stale due to inactivity. Please respond to keep it open.

[beneiltis]

Would you be interessted in an PR with this feature if we take the time? @clusterzx

[BuZZy1337]

If you are using Gemini:
I've built a native Gemini implementation that bypasses the Paperless OCR.
It downloads the actual PDF file and sends it directly to Gemini, allowing it to process tables and complex layouts natively.
It's been running flawlessly in my setup for the last 3 weeks.
You can check it out in PR #864.