MEDIA PREVIEW
Overview & result
Search long PDFs by following a document tree. PageIndex builds the structure; Jev selects sections and candidate pages, then checks whether those pages contain information relevant to the question. Python ranks the paths and returns page references.
From the original post
Inspired by @EGafni’s Twitter thread on combining Jev with PageIndex.
We show how to build long-document search with @typesafeai Jev + PageIndex.
No vector database. No embeddings.
Open source: https://github.com/VectifyAI/jev-doc-search
🧵👇
How Jev fits in the loop
- PageIndex produces a tree of document sections and page ranges.
- Jev Choice narrows the sections and pages; Python keeps several search paths.
- Jev Noul scores candidate pages. The code applies a threshold, with a top-page fallback.
How to reproduce
- Open the author’s repository and install its requirements with pip install -r requirements.txt.
- Set TYPESAFE_API_KEY and PAGEINDEX_API_KEY, then follow the README’s tree_search.py PDF-and-question example.
Why this build matters
Offers a concrete starting point for finding evidence in lengthy reports, with page references and public code.
Limitations
- Not independently run or benchmarked by JevForAgents.
- PDFs are uploaded to PageIndex; API access is required.
- The fallback can return two pages even when none passes the relevance threshold.