How I benchmarked a 100% local RAG pipeline to 9/9 (zero API keys)

Most "chat with your documents" demos work in an afternoon. Then you hit the last So I built a fully local RAG pipeline and, more importantly, a reproducible to prove it actually works. Everything runs on the machine. No Embeddings: Qwen3-Embedding-0.6B (bge-m3 as a fallback) Vector store: Qdrant in