AWS GenAI Certification

AWS GenAI Certification Notes

Study notes for AWS generative-AI material, centered on Amazon Bedrock.

Amazon Bedrock Overview

A serverless API for generative-AI foundation models — one interface to invoke text, chat, image, and embedding models.

• Use pre-built models, fine-tune them on your data, or bring your own.
• First-class support for RAG (Retrieval-Augmented Generation) and agents.
• Serverless — no infrastructure to manage; pay per token / per request.

API Endpoints

Bedrock splits management from inference across four endpoints:

Endpoint	Purpose	Key operations
bedrock	Manage, deploy, fine-tune models	model + provisioning management
bedrock-runtime	Inference — run prompts, generate embeddings	Converse, ConverseStream, InvokeModel, InvokeModelWithResponseStream
bedrock-agent	Manage agents & knowledge bases	agent/KB CRUD
bedrock-agent-runtime	Inference against agents & KBs	InvokeAgent, Retrieve, RetrieveAndGenerate

Converse vs InvokeModel

Prefer the Converse API for chat — it normalizes the message format across model providers, so you can swap models without rewriting request/response shaping. InvokeModel is the lower-level, model-specific call.

IAM

• The root user cannot use Bedrock — use an IAM principal.
• Managed policies: AmazonBedrockFullAccess and AmazonBedrockReadOnly.
• Model access must also be explicitly enabled per-model in the Bedrock console before invocation.

Related

See IAM for the permission model and GraphQL/Lambda for wiring Bedrock into an application backend.