The complete RAG pipeline includes an offline phase (document parsing → text chunking → Embedding vectorization → writing to a vector database) and an online phase (user query → query vectorization → vector retrieval → recalled document reranking → Prompt construction → LLM response generation).