In a retrieval-augmented generation (RAG) design, what role do embeddings and a vector store play?
A.They retrain the foundation model on new documents
RAG avoids retraining entirely — retrieval augments the prompt.
B.Documents are encoded as embeddings; similarity search retrieves relevant chunks to inject into the prompt — grounding answers without retraining
Correct. Embed → store → retrieve → augment is the RAG loop.
C.They compress documents to reduce storage costs
Compression isn't the purpose; embeddings enable semantic retrieval.
D.They translate the prompt into the model's native language
Embeddings capture meaning for similarity, not translation.
Why this answer
RAG grounds generation in retrieved content: docs → embeddings → vector store → similarity search at query time → top chunks inserted into the prompt. The model's weights never change — that's what makes RAG cheap and current.
Correct answer: B · Last reviewed 2026-10-26
Official AWS references
Practice this interactively
Answer without seeing the solution first in the Generative AI Concepts practice session.
Practice Generative AI Concepts