Retrieval-Augmented Generation (RAG) for the AWS AI Practitioner

Retrieval-augmented generation, usually shortened to RAG, is one of the most tested ideas on the AWS Certified AI Practitioner (AIF-C01) exam, because foundation-model applications are its largest domain. If you understand RAG well, a big chunk of the exam becomes easy. Here is the plain-language version.

What RAG actually is

A large language model only knows what it learned during training, so it cannot answer questions about your private or up-to-date data. RAG fixes that by retrieving the relevant documents first, then handing them to the model as context so it can answer grounded in that material. Retrieve, then generate. The model does not need retraining; you just feed it the right information at the moment of the question.

When to use RAG (the exam's favorite question)

• Your answers need current or private data the model was never trained on. RAG, not fine-tuning.
• You need traceability to sources. RAG can cite the retrieved documents.
• The knowledge changes often. RAG updates by changing the documents, with no retraining.
• You need a fixed style or behavior baked into the model itself. That is where fine-tuning fits, not RAG.

RAG vs fine-tuning vs prompting

Prompting is the cheapest lever: shape the input to get a better output. RAG adds grounding in your data without retraining, and is the default when answers must be current, private, or cited. Fine-tuning is the heaviest lever: it adapts the model itself, and you reach for it only when prompting and RAG are not enough. On the exam, the right answer is usually the cheapest lever that meets the requirement.