1 Day 3 Papers
Overview
1Day 3Papers is a fully automated academic newsletter platform that curates, reviews, and delivers research papers daily. It covers three domains — HCI, Digital Health & Medical AI, and Multi-Agent & LLM Agent Systems — serving 40+ subscribers with AI-generated editorial reviews. The entire pipeline runs autonomously: paper collection, AI review generation, admin preview, approval, and subscriber delivery — all without manual intervention.
The Problem
Staying current with academic research is overwhelming. Thousands of papers are published daily across arXiv, PubMed, and DBLP. Researchers and practitioners need a curated, digestible format — not another RSS feed or paper dump.
How It Works
The system operates on a nightly cycle:
• Collection — GitHub Actions fetches papers from DBLP (HCI), PubMed (Digital Health), and arXiv RSS (AI Agents) using keyword filtering and weighted scoring algorithms.
• AI Review Generation — Gemini 2.5 Flash generates structured editorial reviews for each paper: core topic, why it matters, methodology, key findings, and a domain-specific analytical lens (Don Ihde's postphenomenology for HCI, multi-agent collaboration perspective for Agent papers).
• Admin Preview — The system sends preview emails with the actual newsletter design to the admin for quality review before publication.
• Approval & Publish — Upon admin approval, papers are moved from staging to the live archive and deployed to the web.
• Subscriber Delivery — At 7:00 AM KST, Vercel Cron triggers batch email delivery to all subscribers via Resend.
Agent Newsletter: Weighted Scoring System
The AI Agent newsletter uses a custom scoring algorithm to ensure relevant paper selection:
• Multi-Agent / LLM Agent keywords receive weighted scores (title matches count 2x)
• cs.MA category papers get a +3 bonus
• Penalty keywords prevent cross-category contamination
• Final composition: 2 Multi-Agent/LLM papers + 1 Other Language Model (VLM, SLM, World Model) paper
Each review includes a "Multi-Agent Perspective" section that analyzes the paper through the lens of agent collaboration, even for papers that don't directly address multi-agent systems.
Technical Architecture
• Frontend: Next.js 16, Tailwind CSS 4, editorial magazine design
• Email: Resend (batch API), custom HTML templates per newsletter
• Data Pipeline: Python (DBLP, PubMed BioPython, arXiv feedparser)
• AI: Google Gemini 2.5 Flash for review generation
• Automation: GitHub Actions (collection), Vercel Cron (delivery)
• Storage: Redis/Upstash (subscribers), filesystem (archives)
• Deployment: Vercel, GitHub (CI/CD)
Key Design Decisions
• Staging → Approval pipeline prevents low-quality content from reaching subscribers
• Batch email API (Resend) avoids Vercel's 10-second serverless timeout
• KST timezone alignment across Python collection and Node.js delivery
• Retry logic with exponential backoff for Gemini API and Semantic Scholar
Live
→ https://1day3papers.vercel.app