Enforces AWS Well-Architected Generative AI Lens best practices for foundation model workloads on Amazon Bedrock and SageMaker AI. Use when designing GenAI architectures, implementing RAG pipelines, selecting foundation models, configuring Bedrock Guardrails, fine-tuning models, optimizing GenAI costs, securing AI workloads, or applying responsible AI principles including fairness, explainability, and safety.
Enforces AWS Well-Architected Generative AI Lens best practices for foundation model workloads on Amazon Bedrock and SageMaker AI. Use when designing GenAI architectures, implementing RAG pipelines, selecting foundation models, configuring Bedrock Guardrails, fine-tuning models, optimizing GenAI costs, securing AI workloads, or applying responsible AI principles including fairness, explainability, and safety.
Optimize resource efficiency — right-size models; use caching and batching to reduce cost
Establish distributed resilience — design for FM endpoint failures, throttling, and degraded responses
Standardize resource management — centralize model access, prompt catalogs, and configuration
Secure interaction boundaries — validate inputs, filter outputs, enforce least privilege on FM endpoints
Lifecycle Phases
Phase
Key Activities
1. Scoping
Define the business problem; determine if GenAI is the right solution; identify success metrics
2. Model Selection
Evaluate models by modality, accuracy, latency, cost, and context window; benchmark candidates
3. Customization
Prompt engineering → RAG → fine-tuning (progressive investment); evaluate after each stage
4. Development & Integration
Build application logic, implement guardrails, integrate with existing systems
5. Deployment
Deploy to production with monitoring, scaling, and rollback capability
6. Continuous Improvement
Monitor quality metrics; refine prompts, retrieval, and models; manage model lifecycle
Model Selection and Customization Strategy
Decision Tree
Start with prompt engineering (lowest cost, fastest iteration)
↓ Not meeting quality bar?
Add RAG for knowledge-grounded tasks
↓ Still not meeting quality bar?
Fine-tune when behavior/style is the bottleneck
↓ Need maximum control?
Custom training via SageMaker HyperPod
MUST start with prompt engineering and evaluate before investing in RAG or fine-tuning
MUST benchmark multiple models on representative tasks before selecting
SHOULD use the smallest model that meets quality requirements — model right-sizing is the single biggest cost lever
Model Selection Criteria
Criterion
Considerations
Modality
Text, image, audio, video, multi-modal, embedding
Quality
Accuracy on your task benchmarks; reasoning capability
