Launched this week
PyStreamPDF
Intelligence engine for PDFs. Selective retrieval
12 followers
Intelligence engine for PDFs. Selective retrieval
12 followers
Intelligence engine for PDFs. Selective retrieval, structure analysis, token-efficient RAG. 10-50x cost reduction for document processing. - Mullassery/PyStreamPDF
Would love to see built-in OCR support for scanned PDFs so the selective retrieval works on image-based documents too, not just text-extractable ones. That would really widen the use cases.
@zeyneplteklvlu - done. upgrade to the latest version.
Tried it on a 200-page research PDF and was honestly surprised how much the token usage dropped compared to my usual pipeline. The structure analysis picking out section headers cleanly saved me a ton of cleanup work.
the token savings look great but it would be really useful to have a built-in visualization of the extracted document structure, something like an interactive outline tree so you can verify which sections were selected before sending them to the LLM
Would love to see a built-in caching layer for frequently accessed PDFs so repeat queries don't have to reprocess the same documents every time. That could push cost savings even further for teams running the same reports daily.
The selective retrieval approach actually delivers on the token savings, dropped my processing bill noticeably on a stack of dense reports. Took a bit of fiddling to get the structure analysis dialed in for messy scans.
Would love to see a built-in visualization of the document structure it detects, like a tree view of sections and tables after analysis. Right now you have to trust the token savings, but seeing what got parsed and skipped would make debugging retrieval way easier.