// the find
nfpaiva/papervisor
Papervisor is a modular, open-source pipeline for accelerating systematic literature reviews. It combines automation, NLP, and human-in-the-loop screening to help researchers move from raw academic search results to a curated, thematically structured review — fast and reproducibly.
Papervisor is a Flask-based pipeline for managing systematic literature reviews: importing CSV search results, deduplicating them, downloading PDFs, and extracting text for manual review. It's aimed at academic researchers who want an open-source alternative to Rayyan/Covidence and are comfortable running a local Python web app.
The pyproject.toml shows a genuinely serious dev setup for a 2-star repo — mypy strict mode, ruff, bandit, vulture, radon complexity gates, pre-commit, and a CI workflow, which is unusual rigor for a solo research tool. The demo project data (consolidated_papers.csv, extraction_status.json, actual downloaded PDFs) shows the pipeline was tested end-to-end rather than just described. The step-by-step human-in-the-loop design with a real web dashboard (not just CLI scripts) is a sensible fit for the messy, judgment-heavy nature of literature reviews.
Core value prop (LLM-powered screening) is explicitly marked upcoming/WIP, so what's shipped is really CSV import + dedup + PDF download + text extraction — useful plumbing but not the differentiating feature yet. Zero stars/forks and a comparison table against Rayyan and Covidence is a big claim for an alpha-stage single-maintainer project with no docs site despite pyproject referencing readthedocs. The repo ships actual downloaded PDFs and extracted text in the demo/data directory committed to git, which is going to bloat the repo and possibly raises copyright questions for anyone forking it. No obvious deployment story beyond a local shell script — no Docker, no auth on the Flask server, so this is single-user/local-only as-is.