How I Built a Paper Ranking Engine That Saved Me 4 Hours a Week

Disclosure: I work on Paper List, a project on the OpenNomos ecosystem.

Every Sunday, I'd sit down with coffee, open arXiv, and spend hours scrolling through paper titles. 47 papers last week. 3 worth reading. The problem isn't finding papers - it's filtering them. Keyword search returns too much noise. Citation count is a lagging indicator. And conference prestige doesn't tell you if a paper is relevant to your specific project. So I built Paper List.

How It Works

Instead of just matching keywords, Paper List ranks papers by relevance score. You tell it what you're working on, and it learns what matters to you.

The first version was dead simple: a Python script that scraped arXiv, ran embeddings against my project description, and sorted by cosine similarity. 200 lines of code. Ugly as sin. But it worked.

What I Learned

• Relevance scoring beats keyword matching. A paper about "transformer attention mechanisms" might be exactly what you need even if you searched for "NLP efficiency."
• The ranking gets better over time. Every paper you save or dismiss teaches the model. After two weeks, the recommendations were noticeably more targeted.
• Speed matters more than you think. Going from 5 hours to 45 minutes per week isn't just about time saved - it changes your relationship with research. You stop dreading the weekly paper review and start looking forward to what you'll find.

The Numbers

• 47 papers scanned per week → 8 relevant surfaced
• 5 hours → 45 minutes weekly review time
• 3 papers saved per session (up from 1-2 with manual filtering)

It's not magic. It's just a better filter. And sometimes that's all you need.