How to build Karpathy's LLM Wiki. There's a smarter way to use AI with your documents, and it comes from one of the biggest names in the field -- Andrej Karpathy. In this video, I'll show you exactly how to build your own LLM Wiki step by step using Obsidian and Claude Code, so your AI actually remembers and connects everything you feed it.
 
Most AI tools use something called RAG, which means every time you ask a question, the AI searches your documents from scratch. Nothing is saved, nothing builds up. Karpathy's LLM Wiki flips that by having the AI read your documents once and build a structured, interlinked knowledge base out of them. When you add new sources, the wiki grows and gets smarter. In this tutorial, I walk you through the full setup -- creating your Obsidian vault, setting up the folder structure, writing the schema file, ingesting your first documents, and even linting the wiki to keep it healthy. No coding experience needed.
 
Timestamps:
0:00 Intro
1:00 What is RAG and why it falls short
1:51 How the LLM Wiki fixes the problem
2:40 Karpathy's analogy -- Obsidian, LLM, and the wiki
2:59 The three layers of the system
3:54 Setting up Obsidian and the folder structure
5:50 Creating the schema file (CLAUDE.md)
8:03 Installing the Obsidian Web Clipper
8:12 Ingesting your first source document
10:23 Exploring the wiki and graph view
10:42 Adding a second source and watching it update
11:31 Asking cross-source questions
11:57 Linting your wiki for quality
12:52 Use case ideas for students, teachers, and businesses
13:43 Limitations to keep in mind
14:39 Wrap up and next steps
 
Obsidian: https://obsidian.md
CLAUDE.md Schema Template: https://go.teachers.tech/LLM_Wiki_CLAUDE
Claude Code Beginner's Guide: https://youtu.be/s-Mc26Ytz10
 
Subscribe: https://www.youtube.com/@TeachersTech?sub_confirmation=1
Website: https://www.teachers.tech
Instagram: https://www.instagram.com/teacherstechlab
LinkedIn: https://www.linkedin.com/in/teacherstech

YouTube

Andrej Karpathy's LLM Wiki concept addresses a core limitation of RAG-based AI tools: every query starts from scratch with no accumulated knowledge. The solution is to have an AI agent (like Claude Code) read your source documents once and build a persistent, interlinked Markdown wiki in Obsidian. New documents are ingested and integrated into existing pages, contradictions are flagged, and connections are pre-built. The setup involves three layers: raw sources (read-only), the wiki folder, and a schema file (claude.md) that defines rules. A step-by-step walkthrough covers creating the Obsidian vault, folder structure, adding a schema template, ingesting documents, and using a 'lint' command to keep the wiki healthy. Use cases include research, teaching, business knowledge bases, and personal learning.

Karpathy's LLM Wiki - Full Beginner Setup Guide