AIF-C01 Domain 3: Applications of foundation models

RAG, prompt engineering or fine-tuning?

Use prompt engineering when instructions and examples are enough. Use retrieval-augmented generation (RAG) when responses need current or private reference data. Consider fine-tuning when you need repeatable task behaviour or style that prompting alone cannot provide.

• Knowledge Bases for Amazon Bedrock retrieve relevant source chunks for grounded answers.
• Agents for Amazon Bedrock can orchestrate models, data sources and actions.
• Model evaluation should reflect quality, safety, latency and cost for the actual use case.
• Do not send every document in every prompt when retrieval can select relevant context.

Common AIF-C01 decisions

The most capable model is not always the best choice. Compare modality, context needs, accuracy, latency, regional availability, customisation and price. Build a small evaluation set based on real tasks before committing.