[Feature request] OCR Accuracy improvement with AI

[Nolgoshipda]

I would be interested in seeing a feature that utilizes paperless AI to enhance the accuracy of optical character recognition (OCR). While Tesseract OCR is quite effective for printed documents, it still struggles with handwritten documents. The majority of words can be readily guessed based on context and the remaining letters. Given that AI systems are adept at inferring the intended letter and returning desired words, it would be beneficial to have AI systems improving OCR accuracy.

[coolspot18]

This would be a great feature, or the addition of services like Azure Document Intelligence or Google AI. Open AI LLM can also do OCR much better than the built in Paperless models.

[abbottis]

There ist https://github.com/icereed/paperless-gpt, but I',m not pleased with the LLM integration for automatic tagging and title assignment. Here paperless-ai is much more flexible.
It would therefore be great to have the option of integrating external engines (the aforementioned Azure Document Intelligence, Google Document AI, etc.).

[coolspot18]

I installed Paperless-GPT, and it is not as flexible; Paperless-AI is much more polished and would make a wonderful all-in-one solution.

Regarding tagging, I'm still not 100% satisfied with either solution. It seems both create tags that are too granular, resulting in hundreds of individual tags for a repository. It would be great if Paperless-AI could categorize and tag all documents using just a handful of tags instead. I guess each person has different use cases, but I would prefer a common set of tags rather than a laundry list.

[Nolgoshipda]

It seems both create tags that are too granular, resulting in hundreds of individual tags for a repository. It would be great if Paperless-AI could categorize and tag all documents using just a handful of tags instead.

I have the exact same issue. I just turned of the tagging function for now cuz it's to junky rn. We'll have to look for future improvements