The generative AI project lifecycle
Work is structured as selecting a foundation model, adapting it, and deploying at scale using services like SageMaker and Bedrock. Each stage maps to specific AWS tooling and cost tradeoffs.
Generative AI on AWS
3 ideas
Fine-tuning vs. prompting vs. RAG
The book distinguishes adapting model weights, engineering inputs, and retrieval-augmented generation as three levers for task performance. Choice depends on data, cost, and latency constraints.
Generative models as scaled infrastructure
Models are treated as production components needing distributed training, quantization, and cost optimization. Operational concerns dominate over research novelty.