Summary
Wes Roth demonstrates how he built an AI-powered personal knowledge system — a “second brain” — using Claude Code and Obsidian, inspired by Andrej Karpathy’s concept of an LLM wiki. The system automatically ingests information about YouTube videos, AI research papers, sponsor deadlines, and personal notes, organizes it into a linked graph of markdown files, and makes it queryable through the Claude Code desktop app. When asked “what AI topics haven’t I covered that are trending right now?”, the system surfaces answers like Grok, DeepSeek, and Codex by pulling from its own accumulated, self-maintained notes.
The architecture uses Obsidian as the local markdown vault (what Roth calls the “codebase”), Claude as the AI agent that reads, files, cross-links, and summarizes incoming content (the “programmer”), and daily cron jobs that keep the knowledge base updated on autopilot. The system tracks sponsor deadlines, video performance data, and topic clusters autonomously — with Roth describing it as knowing his calendar better than he does. A “doctrine” folder stores high-level strategic insights the AI distills from across the entire vault.
The tutorial covers Obsidian fundamentals — markdown files, double-bracket cross-linking, graph view — and explains how to set up automated ingestion routines. Roth also touches on early experiments with computer use inside Claude Code to automate browser-based tasks like subscription management. The build is designed to be reproducible for creators, researchers, or any knowledge worker who wants a persistent AI librarian organizing information across all their sources.
📺 Source: Wes Roth · Published July 14, 2026
🏷️ Format: Hands On Build
