48.4 Architecture patterns for LLM apps

Overview and links for this section of the guide.

Architecture Patterns

1. Simple Prompt-Response

User → Prompt → Model → Response

For: Simple transformations, one-shot tasks

2. RAG (Retrieval-Augmented Generation)

User → Search → Retrieve Docs → Augment Prompt → Model → Response

For: Q&A over documents, knowledge bases

3. Agent Loop (ReAct)

User → Loop: [Think → Act → Observe] → Response

For: Complex tasks requiring multiple steps

4. Pipeline

Input → Model1 → Model2 → Model3 → Output

For: Classification → extraction → formatting chains

5. Human-in-the-Loop

User → Model → Human Review → Action

For: High-stakes decisions requiring approval

Where to go next