Packing 775 Saved Articles into a 4 MB Vector Store: A RAG Skill That Beats Karpathy’s Wiki
The author built an open‑source RAG skill called chao‑rag‑wiki that compresses 775 collected articles into a 4.8 MB vector index using TurboVec, hybrid dense‑plus‑BM25 retrieval and optional LLM reranking, then compares its zero‑compile latency and full‑recall strengths against Karpathy’s llm‑wiki approach.
