Feature: Chat with pdf/document

[TrendingApp]

Could you add a feature where you can chat with a pdf/document and ask questions related to the pdf/document

[Vali-98]

This is a rather complex feature due to several factors:

1. Document parsing - currently unimplemented
2. RAG - not impossible but does incur some level of calculation + difficulty in storage and searching

I do want this feature too, but it will probably be far off.

[TrendingApp]

Could you make it work with a text file instead of a pdf?

[Vali-98]

No, this will still be a significant amount of work and will not be added soon.

[softpedia-inc]

No, this will still be a significant amount of work and will not be added soon.

When can we expect this feature sir?

[Sphyker789]

No, this will still be a significant amount of work and will not be added soon.

Any updates?

[lrq3000]

Just some ideas after trying manually to do simple RAG like queries:

• For a text file/md file, you can just prepend the document's content in the message, and add the user's prompt at the end after a separation like ------------- . Appending the user's prompt at the end ensures the user's instruction remains fresh in the context window, otherwise it can be forgotten in very long form documents.
• For other types of content, you could use a tool like Pandoc to extract the textual content, then do the same as above to include the user's prompt, so that you offload the conversion work to a 3rd-party tool you don't have to develop. But of course this is going to be prone to a whole host of bugs or documents misreadings...

IMHO text/md file reading would be very useful without being prone to errors nor bugs. But if not implemented, we can already do that just by copy/pasting in the message content.

[bi4key]

Maybe this will be helpfull, to find a concept to start RAG.

New one:
https://k33g.hashnode.dev/how-to-make-an-slm-smarter-short-version

Older one:
https://k33g.hashnode.dev/how-to-make-a-small-language-model-smarter

RAG:
https://k33g.hashnode.dev/understand-rag-with-parakeet

[bi4key]

Or for RAG simply use this app/method:
https://github.com/shubham0204/Android-Document-QA