Amazon Bedrock Popular2023年〜
A fully managed generative AI service providing access to leading foundation models via API
What It Does
Amazon Bedrock is a fully managed service that lets you use foundation models from multiple AI companies - including Amazon, Anthropic, Meta, and Mistral AI - through a unified API. You can integrate generative AI capabilities such as text generation, image generation, embeddings, and chat into your applications without managing any infrastructure. It also supports model customization with your own data (fine-tuning, RAG), enabling you to build AI tailored to your specific use cases.
Use Cases
Used for building automated customer support chatbots, RAG (Retrieval-Augmented Generation) systems powered by internal knowledge bases, automated marketing content generation, code generation and review assistance, document summarization and translation, and creative production through image generation. It covers the full spectrum of generative AI applications.
Everyday Analogy
Think of it like a food court. Opening your own restaurant (AI model) requires massive investment, but at a food court (Bedrock) you can order from multiple famous restaurants (Anthropic, Meta, etc.) all in one place. The menu (API) is standardized, so the ordering process is the same regardless of which restaurant you choose. You can even request custom seasoning (customization) to suit your taste.
What Is Bedrock?
Amazon Bedrock is a generative AI platform service that became generally available in 2023. Previously, using large language models (LLMs) required significant operational overhead including GPU instance procurement, model deployment, and inference endpoint management. Bedrock handles all of this, letting developers access cutting-edge AI models with simple API calls. You can choose between on-demand pricing (pay per use) and Provisioned Throughput for guaranteed capacity.
Knowledge Bases and RAG
Bedrock's Knowledge Bases feature makes it easy to build RAG (Retrieval-Augmented Generation) pipelines that connect your proprietary data (documents, FAQs, manuals, etc.) to foundation models. It automatically chunks and vectorizes documents stored in S3, then stores them in vector stores such as OpenSearch Serverless or Aurora PostgreSQL. When a user asks a question, it retrieves relevant documents and uses them as context for the model's response, enabling accurate answers with reduced hallucination. For detailed coverage of Knowledge Bases and RAG, technical books on Amazon are a valuable resource.
Agents and Guardrails
Bedrock Agents lets you build AI agents that can call external tools and APIs, autonomously completing complex tasks. For example, you can create an agent that searches an inventory system in response to a customer inquiry and processes the order automatically. The Guardrails feature applies safety measures at the API level, including content filtering, PII detection and masking, and topic restrictions, supporting responsible AI operations.
Things to Watch Out For
- Pricing varies by model (input/output token rates), so model selection and cost estimation based on your use case are important
- Generative AI outputs carry the risk of hallucination (generating factually incorrect content), so incorporate human review when using it for critical decisions
