The AWS generative AI stack, end to end, for engineers and architects: Amazon Bedrock for managed models, Amazon SageMaker AI to build your own, and Trainium and Inferentia underneath, plus the retrieval, agent, safety, cost and governance layers that turn a model into a product. A 30-part series that reads from first principles to production. Where it meets vendor-neutral ground it links to the Generative AI guide and the NVIDIA AI guide rather than repeating them.
Series complete · 30 of 30 published
Phase 1 · Platform foundations
- 01What the AWS GenAI Stack Is, End to End
- 02Amazon Bedrock and the Shared Responsibility Model
- 03The Bedrock Model Catalog and Choosing a Model
- 04The Amazon Nova Family
- 05Bedrock vs SageMaker AI
- 06Bedrock Pricing: On-Demand, Provisioned and Batch
- 07Trainium and Inferentia vs GPU
- 08Regions, Quotas and Cross-Region Inference
- 09PrivateLink and VPC Endpoints for Bedrock
- 10Data Residency, KMS and Security
Phase 2 · Calling models, RAG and agents
- 11InvokeModel vs the Converse API
- 12Bedrock Knowledge Bases for RAG
- 13Bedrock Agents
- 14Bedrock Guardrails
- 15Prompt Management, Flows and Caching
- 16Fine-Tuning and Continued Pre-Training on Bedrock
- 17Model Distillation on Bedrock
Phase 3 · Training, data and scale
- 18SageMaker JumpStart
- 19SageMaker HyperPod
- 20Data Prep and Grounding Data
- 21Multi-Agent Collaboration on Bedrock
Phase 4 · Operations, cost and governance
- 22Amazon Q Business and Q Developer
- 23Model Evaluation on Bedrock
- 24Bedrock Data Automation and Multimodal
- 25Observability with CloudWatch and Invocation Logging
- 26Cost Governance and FinOps on AWS
- 27Responsible AI and Watermarking
- 28LLMOps and CI/CD for Bedrock and SageMaker
Phase 5 · Architecture and verdict
DrJha