What it is
Helps researchers re-sort a messy reference library.
What it does: Jev assigns each paper to one of your categories and shows all probabilities for review.
Why it belongs in the showcase: A complete batch-classify loop with versioned taxonomies and a review step.
Caveats: Text-layer PDFs only.
Evidence
Public MIT repo, created 2026-10-20; README in Chinese. Jev classifies with title, abstract and keywords, and adds up to 6,000 characters of body text when there is no abstract. A general LLM can optionally draft 3–5 candidate taxonomies of 2–20 categories from a summary of the selected papers; the user edits and saves one before Jev classifies. Taxonomy edits create versions, results keep model version and input level, and runs export to CSV and JSON.
Caveats
- Relies on the PDF text layer; there is no OCR yet
- Confidence is computed from the probability spread and does not guarantee a correct label
- Auto-drafting taxonomies needs a separate LLM API key
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.