DEMO / RECORDING
Overview & result
Jerry Liu shared DocJev, an open-source Python library, CLI and local app for classifying documents and splitting mixed packets. LiteParse extracts page text locally; Jev predicts document categories and packet boundaries. LlamaParse is an optional OCR path for difficult inputs. The project is independent of LlamaIndex hosted Classify and Split APIs.
From the original post
DocJev is the fastest way to classify and split complex document packets ⚡️. (and the default is fully free and OSS!)
I made a sick teaser video below. TY Opus 5.5 🙏
check it out: https://github.com/jerryjliu/docjev
How Jev fits in the loop
- Extract page text from PDF, DOCX or PPTX with LiteParse.
- Give DocJev natural-language category rules. Jev predicts a category or the boundaries between documents in a packet.
- Review the ordered page ranges and optionally export each split PDF.
How to reproduce
- Clone https://github.com/jerryjliu/docjev and run uv sync.
- Set TYPESAFE_API_KEY and run uv run docjev doctor --smoke.
- Run the classify and split commands from the repository README on its public example documents.
Why this build matters
The repository provides inspectable source code, examples and a visual benchmark report for document classification and packet splitting.
Limitations
- Jev inference uses a hosted service and requires a TypeSafe API key.
- DOCX and PPTX inputs require LibreOffice; optional cloud OCR requires separate credentials.