Karpathy—a key figure in modern AI development—operates his personal knowledge system using just three folders rather than complex tools like Notion or Obsidian.
System Architecture
The setup involves creating three subfolders within a project directory:
- raw/ — repository for articles, notes, screenshots
- wiki/ — AI-organized useful knowledge
- outputs/ — generated answers, reports, research
This structure separates collection from organization, treating them as distinct processes that shouldn't occur simultaneously.
Schema Configuration
A CLAUDE.md file serves as the standing instruction set, kept "super simple and flat" according to Karpathy. Rather than code or databases, it's plain text specifying what matters, how to organize it, and procedures for new material.
Compounding Knowledge
With ten-plus wiki articles, cross-source queries become possible. Questions about client patterns, proposal accuracy, or competitive analysis draw from the entire collection. Each inquiry and saved answer strengthens subsequent decision-making.
Business Application
Most small business owners possess scattered institutional knowledge across presentations, emails, and personal notes. This system transforms fragmented insights into scalable infrastructure accessible to any team member within seconds, automatically cross-referenced and continuously updated.
Pixelworkers implements this infrastructure for small and mid-sized businesses, configuring agents, writing schemas, and establishing collection pipelines that improve with each use.
