Show HN: Mcptube – Karpathy's LLM Wiki idea applied to YouTube videos (github.com)

Mcptube-vision, an advanced YouTube video knowledge engine, has been launched, building on the principles of Karpathy's LLM Wiki. It transforms YouTube videos into a structured knowledge base by leveraging both transcripts and visual frame analysis. Unlike traditional video analysis tools that reconstruct knowledge afresh with each query, mcptube-vision compiles information from every video ingested, creating a persistent wiki that becomes more intelligent with each addition. This system reflects a significant advancement in video knowledge processing, enabling a continuous learning and synthesis of concepts, rather than isolating knowledge. The mcptube-vision architecture is composed of distinct subsystems that collectively enhance the knowledge retrieval process. The ingestion pipeline extracts metadata, scenes, and descriptions, while the WikiEngine merges new video contributions into an evolving knowledge base. Key technical innovations include the use of scene-change detection for extracting frames with higher information density and an FTS5 index that allows for sub-millisecond keyword retrieval, making searches efficient and immediate. This design not only facilitates knowledge compounding, where new insights augment existing ones, but also allows for an auditable and transparent update mechanism, ensuring that the system continually refines its understanding of topics and concepts derived from video content.