I kept seeing RAG tutorials that stop at “vector DB + prompt” and break down in real systems. I put together a roadmap that reflects how modern AI search actually works: – semantic + hybrid retrieval (sparse + dense) – explicit reranking layers – query…