jev-information-extraction
PDF question-answering demo where Jev chooses which extracted text chunk answers each question, shown with its location.
Overview
Jev Information Extraction lets you ask questions about a PDF, such as 'What is the GST number?' or 'What is the total invoice amount?', and uses Jev to rank the source text that answers them. Text chunks and positions are extracted locally with pdf-inspector and PDFium renders the page, without calling Jev. For each question, the backend creates a Choice whose options are the page's chunks, then sends the page text and all questions for that page in one system_one call.
Returned probabilities map back to chunks; the UI shows the top chunk and an expandable second-best match, with red boxes marking them on the page for inspection. The README is clear that this is question-to-chunk selection per page, not a benchmark, and the app does not compute accuracy or F1. A FastAPI service runs evaluation and serves the frontend.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which text chunk on this page answers the question? | — |
How it works
- Extract text chunks and positions from the PDF
- Build one Choice per question over the page's chunks
- Call Jev with page text and questions
- Show top matches highlighted on the page
Strengths
- Answers point to exact source text and location
- No generated text to verify
- Multiple questions per page in one call
Caveats
- Not a benchmark; no accuracy reported
- Per-page selection only
At a glance
- Type
- Demo
- Status
- prototype
- Language
- Python
- Runtime
- FastAPI backend with web UI
- Has code
- yes
Use cases and integrations
How curated lists describe it
Ask questions about a PDF. Use Jev to rank the source text that answers them. Inspect each match, its probability, and its location on the original page.
logicrw/awesome-jev-projectsParsing the PDF and extracting the relevant information.
hellogumbo/awesome-jev
Frequently asked questions
What does Jev decide?
Which extracted chunk on a page answers each question.
Does it generate answers?
No, it selects existing text from the PDF.
Is accuracy measured?
No, the README says it is a demo, not a benchmark.
Related projects
Long-document search that walks a PageIndex tree, asking Jev Choice which page or branch answers the question, with no embeddings.
ProjectprototypePythonClassifies tax PDF pages into 261 IRS forms and 7 page kinds with one Jev request per page and a confidence gate.
ProjectproductionTypeScriptRust tool that asks Jev which PDF pages genuinely need OCR, extracting text pages locally and sending only the rest to OCR.
ProjectbetaRustLocal library that turns an image into a transparent text state (OCR, boxes, labels) so Jev can decide over screenshots.
ProjectbetaPythonWindows office workbench with search, OCR and batch renaming, where Jev optionally helps pick extracted contract fields.
ProjectbetaRustDocker service that classifies the Paperless-ngx inbox with Jev (or local Clef via Ollama), setting confident fields and queueing the rest.
Projectbeta