When designing a Generative AI solution, your choice of architecture pattern depends on your data and context needs. The most common patterns are:

• Direct FM invocation: The application calls a foundation model directly with a prompt, without retrieving external context.
• Retrieval-Augmented Generation (RAG): The application retrieves relevant documents or chunks from your data (e.g., from S3 and a vector store) and injects them into the prompt.
• Agent or tool-calling pattern: The model can call APIs, databases, or tools to perform actions or gather more data.
• Multimodal pattern: The application works with more than one modality, such as text combined with images or audio.