Retrieval heat 1 · Quiet★ 6heat incomplete (no X)

jev-doc-search

Long-document page search via PageIndex tree + Jev Choice (optional Noul check); no vector DB.

What it is

Long-document search that turns “which page answers this?” into Jev Choice calls, with PageIndex building a hierarchical tree so each choice stays small.

What it does: Flat page search for short PDFs; tree beam for 10-K-scale reports; optional Noul check on the retrieved page.

Why it belongs in the showcase: Clear hierarchical beam + passage gate without a vector DB, with public financial-report walkthroughs.

Caveats: Dual API keys; author numbers are not SLAs; hosted model sees document text.

Evidence

Public Apache-2.0 README with NVIDIA/Citigroup 10-K walkthrough; PageIndex tree + Jev Choice page search.

Caveats

  • Needs TYPESAFE_API_KEY + PAGEINDEX_API_KEY
  • Author-reported table is not an SLA
  • Flat Choice hits token/option limits on long PDFs — tree path required

Unofficial case page. Stars / list position observed for heat; do not treat heat as a quality SLA. Review the upstream README before you depend on it.