In this video, I’m showing you a simple way to turn your YouTube backlog into an actual knowledge system you can query—without doing any manual relationship building. I’m looking at 36 of my most recent videos mapped into a graph of nodes, where each dot is a video with tags, the video link, the raw transcript file, and a clean explanation of what it’s about plus the key takeaways. The best part is the backlinks: I can jump between concepts like the WAT framework, Claude Code, bypass permissions mode, human review checkpoints, and tools I’ve mentioned like Perplexity, Visual Studio Code, Nano Banana, and n8n. The workflow is basically what Andrej Karpathy has been talking about with LLM knowledge bases: data ingest (source docs like PDFs or transcripts), then letting an LLM organize everything into a usable structure. I’m doing this with Claude Code and viewing it inside Obsidian, which just makes it easy to browse the markdown files visually. In practice, I can tell Claude Code to grab transcripts from recent videos and organize them into a wiki-style folder structure—and it just figures out the relationships on its own. You can set this up in about 5 minutes, and once it starts filling up, the patterns across tools, skills, and MCP servers become insanely easy to explore.