How RAG grounds LLM answers in your data: chunking, embeddings, hybrid BM25 search, reranking, faithfulness, and RAG vs fine-tuning vs long context.