Jev Information Extraction
Find the PDF text chunks most relevant to a question.
What Jev decides
Which extracted text chunk on a page best answers the question.
Jev Information Extraction ranks extracted PDF chunks with Jev and marks selected passages on the original page.
- Jev role
- Rank source passages as candidate answers to a PDF question.
- Input
- The question, page text, and extracted page-local chunks.
- Output
- Choice probabilities mapped back to original PDF text locations.
- Decision frequency
- Once per evaluated page, batching its questions.
- Fallback
- The UI preserves original extracted text for inspection; scanned pages need a text layer.
How it works
Overview
The backend extracts embedded PDF text and positions. For each question and page, it gives Jev a Choice over page-local chunks, then displays the highest-ranked source text, an alternative, probabilities, and boxes on the page preview. It does not normalize fields, join values across chunks, run OCR on scanned pages, or measure extraction accuracy. Uploaded documents stay in server process memory; the README describes this as a demo rather than a private multi-user service.
- Language
- Python
- Author
- abhishekmamdapure
Sources
Related projects
Notra
Use narrow Jev judgments for brand sentiment and list position inside a broader AI visibility platform.
What Jev decides
Positive, neutral, or negative sentiment and list position from one through ten or none.
typesafe-computer-use
A native macOS computer-use loop built from local OCR, accessibility data, and bounded Jev actions.
What Jev decides
Which computer action, visible item, site, or off-screen accessibility control to use next.
Jev Chat Assistant
An Android conversation copilot that uses Jev to interpret chats and rank reply candidates without sending for the user.
What Jev decides
The likely intent, conversation risk, reply need, best next action, unresolved tension, and which of three candidate replies fits best.