ziwenhahaha/daily-paper-reader

What it solves

Daily Paper Reader provides a centralized, automated system for researchers to discover and read AI papers without needing to manually browse multiple repositories. It solves the problem of information overload by creating a personalized feed of new research from sources like arXiv and OpenReview, while integrating reading and analysis tools into a single interface.

How it works

The project uses a serverless architecture leveraging GitHub Actions for backend processing and GitHub Pages for the frontend. It automatically fetches new papers based on user-defined keywords and research interests. The system employs a multi-stage pipeline involving semantic recall (using embedding models like BGE), reranking (using models like Qwen-Reranker), and LLM-based refinement (via DeepSeek) to filter and summarize papers. It also includes a PDF processing layer to extract figures and provide AI-powered Q&A for the papers.

Who it’s for

• Individual Researchers: Those who need to stay current with their specific research niche.
• Academic Labs: Teams looking to create a shared repository of tracked papers and reading results.
• Students: Anyone needing a streamlined workflow for discovering, summarizing, and analyzing academic papers.

Highlights

• Zero-Server Deployment: Runs entirely on GitHub Actions and Pages, requiring no dedicated server.
• Personalized Feed: Customizes recommendations based on keywords, research directions, and intent queries.
• AI-Enhanced Reading: Combines summaries, original text, and an AI chat interface for in-context questioning.
• Multi-Source Integration: Supports arXiv, OpenReview, bioRxiv, medRxiv, and ChemRxiv, as well as major AI conferences (e.g., NeurIPS, ICLR, CVPR).
• Automated Figure Extraction: Detects and crops figures from PDFs for better visual browsing.