A company wants the LLM to answer questions using ONLY its internal policy documents — preventing invented answers. Which technique grounds the model on company data?
A.Fine-tuning on the documents
Fine-tuning adapts behavior/style — it won't reliably inject current facts.
B.Retrieval-augmented generation (RAG) — retrieve relevant document chunks and include them in the prompt
Retrieve-then-generate = grounded answers. Correct.
C.Increase the temperature setting
Higher temperature increases randomness — opposite of factual grounding.
D.Use a larger foundation model
Model size doesn't grant access to your private documents.
Why this answer
RAG retrieves relevant content from a knowledge source (e.g., Azure AI Search over the policy docs) and injects it into the prompt so the model answers from real data — the standard 'chat on your data' pattern. Fine-tuning teaches style/behavior, not factual grounding.
Correct answer: B · Last reviewed 2026-02-14
Official AWS references
Practice this interactively
Answer without seeing the solution first in the Generative AI practice session.
Practice Generative AI