MLA-C01 Practice Question 76: Knowledge Base RAG
AWS Certified Machine Learning Engineer – Associate (MLA-C01) · Test 8 · question 6 of 10
A company is building an internal assistant on Amazon Bedrock that answers employee questions about HR policies. About 2,000 policy documents are stored in Amazon S3, and HR edits some of them every week. Answers must reflect a policy change within a day, and each answer must cite the documents it was based on so employees can check it. The team has no labeled question-and-answer data and does not want to train or maintain a custom model.
Which approach should the team use?
Show answer and explanation
Answer:
Amazon Bedrock Knowledge Bases implements retrieval-augmented generation (RAG). It ingests documents from a data source such as S3, converts them into vector embeddings, retrieves the passages relevant to each question, and passes them to the model. Responses can include citations back to the source documents. When policies change, syncing the data source re-ingests them, so answers reflect updates without training anything (option D).
- A is wrong because fine-tuning bakes knowledge into model weights. Weekly retraining costs time and money, still lags behind edits, cannot cite the document an answer came from, and is the custom model maintenance the team ruled out. - B is wrong because 2,000 documents will not fit in a model's context window, and even a fraction of them in every request would make each call slow and expensive. - C is wrong because distillation makes a smaller model imitate a larger one to cut cost and latency. It does not give either model access to current policy content or produce citations.
Reference: Retrieve data and generate AI responses with Amazon Bedrock Knowledge Bases