Understand Knowledge

Takes a Karpathy-pattern wiki (the three-layer setup with raw sources, markdown files with wikilinks, and index.md) and builds an interactive knowledge graph from it. Extracts explicit structure like categories and wikilinks automatically, then uses LLM agents to find implicit relationships between articles that aren't linked but should be. The output is a force-directed graph with entities, claims, and topics that gets auto-launched in a dashboard. Useful if you're maintaining one of these wikis and want to see the conceptual structure beyond just the file tree. The analysis is batched and runs concurrently, so it scales reasonably to larger knowledge bases.