I Built a Self-Updating SEO Brain Inspired by Andrej Karpathy's LLM Wiki

Short summary

Andrej Karpathy's LLM Wiki pattern addresses a fundamental RAG limitation: systems that rediscover knowledge from scratch on every query. This post implements that pattern for SEO monitoring using a 3-layer architecture (immutable raw data → LLM-maintained markdown wiki → schema rules) with daily agents that track keyword trends, identify issues, and connect code changes directly to metrics. The system replaces broken Neo4j/ChromaDB setups with simpler markdown, enabling rich historical context and causal-chain analysis through Obsidian's graph visualization.

• •Implements Karpathy's LLM Wiki pattern to solve RAG's knowledge-rediscovery limitation with a 3-layer markdown architecture
• •Daily automation agents ingest GSC data, page audits, and commits, updating wiki with cross-links and causal chains tracking issue→fix→metric improvements
• •Demonstrates real production impact: 81% click improvement, eliminating Neo4j/ChromaDB operational overhead

Generated with AI, which can make mistakes.