Core Concepts
- Amazon Bedrock is a fully managed service to access foundation models via API.
- Available model providers: Amazon (Titan), Anthropic (Claude), AI21 Labs, Cohere, Meta (Llama), Stability AI.
- No infrastructure to manage — serverless, pay-per-use pricing.
- Models are accessed via the Bedrock API; your data is not used to train base models.
- Supports text generation, image generation, embeddings, and chat.
Key Features
- Model Playground: test and compare different models interactively.
- Knowledge Bases: connect FMs to your data sources for RAG.
- Agents: build autonomous agents that can plan, execute, and orchestrate tasks.
- Guardrails: configure content filters, denied topics, and PII handling.
- Fine-tuning: customize models with your own labeled data (continued pre-training or instruction tuning).
- Model Evaluation: compare model outputs using automatic or human evaluation.
Model Selection
- Amazon Titan: general-purpose text and embeddings, good for enterprise use.
- Anthropic Claude: strong at analysis, coding, and following instructions.
- Meta Llama: open-weight model family for flexible deployment.
- Stability AI: specialized in image generation (Stable Diffusion).
- Choose based on task, latency, cost, and accuracy requirements.
Security
- Data is encrypted in transit and at rest.
- Your data is not shared with model providers and not used to improve base models.
- IAM policies control access to models and features.
- VPC endpoints available for private connectivity.
- CloudTrail logs all Bedrock API calls for auditing.
Practice Bedrock Questions
Put your knowledge to the test with practice questions.