Machine Learning
Amazon Bedrock
Amazon Bedrock provides managed access to foundation models and capabilities for building generative AI applications. It centralizes model invocation, inference profiles, customization, evaluation, safety, agents, and retrieval without customers managing model servers.
Key points
- Converse offers a common message interface across supported models
- On-demand and provisioned modes address different traffic economics
- Model availability, features, quotas, and data residency vary by model and Region
- Intelligent prompt routing and cross-Region inference solve different routing problems: model selection versus serving-location capacity
- Supported models can use structured outputs and prompt caching, each with API, model, and inference-mode constraints
- Model invocation logging is opt-in per account and Region and can deliver sensitive request, response, and usage evidence to CloudWatch Logs or S3
When to use it
- Build a multi-model enterprise assistant
- Generate and classify content through governed APIs
Exam tips
- Choose Bedrock for managed foundation models; use SageMaker AI when custom hosting and endpoint control dominate
- Evaluate each candidate on representative quality, latency, safety, and cost rather than model size alone
- A valid structured output is not proof of authorization, factuality, or compliance with business rules