PeerLens

Open infrastructure for transparent research quality signals.

PeerLens helps researchers, reviewers, and institutions assess papers with explainable automated checks — not a single opaque score. It fetches metadata from open sources (Crossref, arXiv), runs pluggable quality signal checkers, and returns a structured report you can inspect, extend, and fork.

Decision support for science, not a proprietary credit rating.

Preview

What it does today (v0.3)

• Ingest papers by DOI or arXiv ID (URL formats supported)
• Auto-fetch arXiv PDFs and extract section headings (methods, results, etc.)
• Upload PDFs for full-text artifact and section analysis
• Run automated quality signals: metadata completeness, Crossref retraction checks, code/data artifact links
• RAG Q&A — ask questions about a paper with cited excerpts (POST /api/v1/papers/ask)
• Expose a FastAPI service, CLI, and web UI

Quick start

Requirements

• Python 3.11+
• pip (or uv)

Install

git clone https://github.com/LordKay-sudo/peerlens.git
cd peerlens
python -m venv .venv

# Windows
.venv\Scripts\activate

# macOS / Linux
source .venv/bin/activate

pip install -e ".[dev]"
cp .env.example .env

Run the API

peerlens serve --reload

Open http://localhost:8000/docs for interactive API docs.

Run the web UI

In a second terminal:

cd web
npm install
npm run dev

Open http://localhost:3000. The frontend proxies /api/* to the FastAPI backend on port 8000.

Analyze a paper (CLI)

peerlens analyze 10.1038/nature12373
peerlens analyze 2301.07041

Ask a question (HTTP)

Requires OPENAI_API_KEY in .env.

curl -X POST http://localhost:8000/api/v1/papers/ask \
  -H "Content-Type: application/json" \
  -d '{"identifier": "2301.07041", "question": "What methods does this paper use?"}'

After analyzing a paper in the web UI, use the Ask this paper panel below the report.

Architecture

peerlens/
├── src/peerlens/     # FastAPI backend
│   ├── api/              # FastAPI routes
│   ├── services/
│   │   ├── ingestion/    # DOI (Crossref), arXiv metadata fetchers
│   │   ├── signals/      # Pluggable quality checkers
│   │   └── reports.py    # Orchestrates ingest → signals → report
│   └── models/           # Pydantic schemas
└── web/              # Next.js frontend (React, Tailwind, Framer Motion)

Signal checkers implement a small interface (SignalChecker) and return structured QualitySignal objects — each with an ID, severity (info / warning / concern), message, and optional evidence span.

Roadmap

• PDF ingestion and section-aware extraction
• RAG over paper + references
• LLM-assisted claim / method extraction (opt-in, traced)
• Human review rubric API (blinded expert scores)
• Evaluation harness on known retractions / replication failures
• Plugins for Zotero, OpenReview, institutional repos

Limitations

PeerLens v0.1 analyzes metadata only (title, abstract, authors, dates). Absence of a warning is not endorsement. Automated signals are heuristics — always pair with expert judgment for high-stakes decisions.

Contributing

Issues and PRs welcome. See GitHub Issues for planned work.

License

MIT — see LICENSE.

Author

Built by LordKay.