Main Concept
Amazon Bedrock is a fully managed AWS service that is used to build Generative AI applications, providing access to several Foundation Models from different providers.
Key Points
- All the models are accesible through a single API
- You keep control of your data used to train the model.
- Pay-per-use pricing model
- It provides security, privacy, governance and responsible AI policies.
- You can evaluate models and customize with your own data.
- It’s serverless, so you don’t have infrastructure to manage, and you can use familiar AWS services to integrate to applications.
Main Features
Security and Privacy
- Your data is encrypted at rest in the same AWS Region and is not shared outside of it.
- It supports compliance standards like SOC, ISO, HIPPA and GDRP
- You can use AWS Private Links to connect resources on VPCs to the Amazon Bedrock endpoints.
Guardrails
- You can implement guardrails via corporate policies to configure things such as:
- Filters for inappropriate content
- Monitor violations
- Redact Personal Identifiable Information (PII)
Agents
- With Amazon Bedrock agents you can run multi-step tasks and interact with other AWS Services and Data Source (like S3 or DynamoDB, etc..
- With agents you can automate task like API Calls, Engineering Prompts, and general Orchestration of you Gen AI pipelines.
Amazon Bedrock General Diagram
Related Notes
- Amazon Bedrock Foundation Models
- Model evaluation with Amazon Bedrock
- Model Customization with Amazon Bedrock
- Fine-tunning in Amazon Bedrock