Updated June 2026

Generative AI on AWS with Amazon Bedrock

Class Duration

21 hours of live training delivered over 3 days.

Student Prerequisites

Professional software development experience (Python is helpful)
Familiarity with REST APIs and cloud fundamentals
An AWS account with access to Amazon Bedrock

Target Audience

Developers, solution architects, and platform teams building generative AI applications and agents on AWS with Amazon Bedrock. Relevant for teams adding AI capabilities to existing AWS workloads, and for organizations standardizing how they build, secure, and govern generative AI on AWS. This is the AWS counterpart to the Azure AI Foundry course. For general AWS infrastructure foundations, see Architecting on AWS Essentials and AWS Cloud Foundations.

Description

Amazon Bedrock is AWS's managed platform for building generative AI: a broad foundation-model catalog plus the managed services around it for retrieval, agents, safety, orchestration, and evaluation. This three-day course teaches building production generative AI on AWS end to end. Participants start with model selection across the Bedrock catalog (Anthropic Claude, Amazon Nova, Meta Llama, and others), then prompt engineering, structured outputs, and tool use. The course moves into retrieval-augmented generation with Knowledge Bases for Amazon Bedrock, agents with Bedrock Agents and the AgentCore runtime, Bedrock Guardrails for content safety and grounding, and Bedrock Flows for orchestration. The final day covers model customization through fine-tuning, distillation, and Nova Forge, systematic evaluation, where Amazon Q fits for developers and the business, and deploying generative AI on AWS with sound identity, security, cost, and observability controls. Labs build a realistic Bedrock application and agent throughout.

Learning Outcomes

Navigate Amazon Bedrock: the foundation-model catalog, the playground, and the core APIs.
Select and compare models (Anthropic Claude, Amazon Nova, Meta Llama, and others) for cost, quality, and latency.
Build prompt pipelines with structured outputs and tool use on Bedrock.
Implement retrieval-augmented generation with Knowledge Bases for Amazon Bedrock.
Build agents with Bedrock Agents and the AgentCore runtime, connecting tools and knowledge.
Apply Bedrock Guardrails for content safety, grounding, and sensitive-data protection.
Customize models through fine-tuning, distillation, and Nova Forge, and evaluate quality systematically.
Deploy generative AI on AWS with sound identity, security, cost, and observability controls.

Training Materials

Comprehensive courseware is distributed online at the start of class. All students receive a downloadable MP4 recording of the training.

Software Requirements

An AWS account with Amazon Bedrock model access enabled, the AWS CLI with an IAM role granting appropriate permissions, Python 3.12+ (Boto3) and/or Node.js 22+, a vector store for Knowledge Bases (such as Amazon OpenSearch Serverless), and Git.

Training Topics

Generative AI on AWS in 2026

Amazon Bedrock as AWS's managed generative AI platform
The services around the models: RAG, agents, guardrails, flows, and evaluation
Reference architectures for generative AI on AWS
How Bedrock relates to SageMaker and Amazon Q

The Bedrock Model Catalog

Foundation models from Anthropic, Amazon, Meta, Mistral, Cohere, and others
Amazon Nova and Titan model families
The Bedrock Marketplace and specialized models
Choosing a model for cost, quality, and latency

Invoking Models and Prompt Engineering

The Invoke and Converse APIs
Streaming responses
Prompt engineering patterns on Bedrock
Prompt management and versioning

Structured Outputs and Tool Use

Structured and JSON output
Tool use and function calling
Multi-step tool pipelines
Validation and error recovery

Retrieval-Augmented Generation with Knowledge Bases

Knowledge Bases as a managed RAG pipeline
Document ingestion, chunking, and embeddings
Vector stores and retrieval configuration
Citations and grounding quality

Bedrock Agents

Defining agents, instructions, and action groups
Connecting tools and Knowledge Bases
Orchestration and reasoning traces
Testing and iterating on agents

AgentCore and the Agent Runtime

AgentCore as a microVM-based agent platform
Running agents in isolated, production-grade environments
Memory, identity, and tool gateways
Moving from prototype agents to production

Bedrock Flows and Orchestration

Visual orchestration of prompts, models, and logic
Composing knowledge bases, agents, and guardrails into flows
Conditional and multi-step workflows
Versioning and deploying flows

Guardrails for Bedrock

Content filters and denied topics
Grounding and hallucination checks
Sensitive-information and PII protection
Associating guardrails with agents, knowledge bases, and flows

Model Customization

Fine-tuning and continued pre-training
Model distillation for cost and latency
Customizing Amazon Nova with Nova Forge
Managing custom models and provisioned throughput

Evaluation and Quality

Model and application evaluation on Bedrock
Automated and human-in-the-loop evaluation
Building evaluation datasets and scorers
Gating releases on quality results

Amazon Q on AWS

Amazon Q Developer for building and operating on AWS
Amazon Q Business for enterprise knowledge
How Amazon Q builds on Bedrock
When to use Q versus a custom Bedrock application

Security, Cost, and Operations

IAM, identity, and access for Bedrock
Data privacy, residency, and VPC integration
Observability, logging, and tracing
Cost tracking and model-selection controls