AWS Certified Generative AI Developer – Professional

GenAI Foundations & Amazon Bedrock Core

• 1. GenAI landscape: LLMs, Multimodal, Embedding & Diffusion models — how they differ from traditional ML
• 2. Amazon Bedrock: model catalog, API (InvokeModel / Converse / Streaming), Cross-Region Inference
• 3. Amazon Nova family (Nova Micro, Lite, Pro, Premier, Canvas, Reel) — capabilities & use cases
• 4. Amazon Bedrock Marketplace: 3rd-party model discovery, subscription, and deployment
• 5. Provisioned Throughput vs On-Demand vs Batch Inference — pricing and trade-offs
• 6. Hands-on: Implementing Cross-Region Inference with Amazon Bedrock
• 7. Hands-on: Using FMs Stable Diffusion in AWS Bedrock to Generate Images

Prompt Engineering & Bedrock Prompt Technical

• 1. Prompt engineering techniques: zero-shot, few-shot, chain-of-thought, ReAct — when to use each
• 2. Amazon Bedrock Prompt Management: versioned templates, parameterization, approval workflows
• 3. Amazon Bedrock Prompt Flows: visual no-code builder, conditional branching, reusable nodes
• 4. Prompt Caching on Bedrock — reduce latency & cost for repeated context blocks
• 5. Output control: JSON Schema enforcement, structured outputs, temperature / top-p / top-k tuning
• 6. Hands-on: Implement Amazon Bedrock Prompt Caching with Amazon Nova

Data Governance for GenAI

• 1. Amazon Glue & DataBrew
• 2. AWS Lake Formation
• 3. Amazon S3 Data Lake
• 4. Amazon SageMaker Data Wrangler
• 5. Amazon Bedrock Data Automation
• 6. Amazon DataZone
• 7. Hand-on: Document Processing Using Amazon Bedrock Data Automation (BDA)
• 8. Hands-on: Implement solution table-level access control on data lake tables using AWS Glue 5.0 with AWS Lake Formation
• 9. Hands-on: Advance End-to-end feature engineering data with Glue DataBrew

Building Enterprise RAG Systems

• 1. RAG architecture deep-dive: chunking strategies, embedding models (Amazon Titan v2, Cohere), hybrid search
• 2. Amazon Bedrock Knowledge Bases: managed ingestion, OpenSearch Serverless, Aurora pgvector, Amazon S3 Vectors (new)
• 3. Amazon Bedrock Reranker & query transformation (expansion, decomposition) for retrieval quality
• 4. Model Context Protocol (MCP): architecture, client/server patterns, integration with Knowledge Bases
• 5. GraphRAG on Amazon Neptune — entity-relationship aware retrieval for complex knowledge graphs
• 6. Hands-on: Building multimodal RAG applications using Bedrock KB with Multimodal Embeddings (PDF,Word,Excel,PowerPoint) and S3 Vectors.
• 7. Hands-on: Deep dive Model Context Protocol on AWS

AI Safety, Guardrails & Evaluation, Responsible AI

• 1. Amazon Bedrock Guardrails: content filters, topic denial, word filters, PII redaction, grounding checks
• 2. Prompt injection & jailbreak defense: input sanitization, safety classifiers, adversarial testing
• 3. Hallucination reduction: grounded generation, confidence scoring, JSON Schema enforcement
• 4. Data privacy controls: Amazon Macie, Comprehend PII, VPC Endpoints, Bedrock PrivateLink
• 5. AI Governance & compliance: Model Cards, Glue data lineage, CloudTrail audit, bias drift monitoring
• 6. Observability stack: CloudWatch (token usage, hallucination rates), X-Ray tracing, Bedrock Model Invocation Logs
• 7. Hands-on: Configure multi-layer Guardrails (PII + grounding + content + Injection/Jailbreak + Grounding) on a live Bedrock Agent
• 8. Hands-on: Build GenAI Applications Using Amazon Bedrock With AWS PrivateLink To Protect Your Data Privacy

Agentic AI, Amazon Bedrock Agents & AgentCore, Multi-Agent Systems

• 1. Agentic AI fundamentals: memory, planning, tool use, multi-step reasoning (ReAct / chain-of-thought)
• 2. Amazon Bedrock Agents: Action Groups, inline agents, supervisor patterns, alias management
• 3. Amazon Bedrock AgentCore (2025): managed runtime for long-running agents — memory, sessions, identity
• 4. AWS Strands Agents SDK & AWS Agent Squad — multi-agent orchestration on AWS
• 5. MCP Servers on Lambda / ECS
• 6. Hands-on: Scaling content review operations with multi-agent workflow
• 7. Hands-on: High level design multi-agent orchestration using Amazon Bedrock

FM Customization, Fine-Tuning & Amazon Kiro

• 1. Customization spectrum: prompt engineering → RAG → fine-tuning → continued pre-training
• 2. Amazon Bedrock Fine-Tuning & Model Distillation: dataset format, LoRA/PEFT, job configuration
• 3. Amazon Kiro (2025): AI-native IDE for spec-driven development — spec sheets, auto hook generation, agent steering
• 4. Amazon Q Developer integration: inline code suggestions, /dev agent, security scans, unit test gen
• 5. CI/CD for GenAI with AWS CodePipeline: automated prompt regression tests, model rollback, GenAI gateway pattern
• 6. Hands-on: Fine-tune Amazon Nova models using your own data to create specialized models tailored to your specific use cases.

SageMaker Unified Studio & MLOps for GenAI

• 1. Amazon SageMaker Unified Studio (2025): unified data + ML + GenAI workspace, project catalog, governance
• 2. SageMaker Lakehouse integration: unified data access across S3, Redshift, Glue for GenAI pipelines
• 3. SageMaker HyperPod & training best practices for large FM fine-tuning jobs
• 4. Model deployment strategies: SageMaker endpoints, Bedrock custom model import, traffic shifting
• 5. MLOps patterns: SageMaker Pipelines, Model Registry, A/B testing, canary deployments for FMs
• 6. Hands-on: Develop and deploy a generative AI application using Amazon SageMaker Unified Studio
• 7. Hands-on: Sử dụng Inference Components (ICs) để host nhiều LLMs trên SageMaker Real-Time Endpoints

Orchestrate Generative AI Workflows

• 1. Amazon API Gateway
• 2. Amazon EventBridge
• 3. Amazon SNS/SQS
• 4. Amazon Amplify
• 5. Amazon Cognito
• 6. AWS Step Functions
• 7. Hands-on: Amazon Bedrock AgentCore Evaluations: LLM-as-a-Judge in Production
• 8. Hands-on: Build solution Generative AI to the data warehouse with Amazon Bedrock and Amazon Redshift

End-to-end: Building GenAI applications with AI Agents that reason, decide and act independently.

• Scenario: Building GenAI applications with AI Agents that reason, decide and act in production
• 1. Designing and implementing AI agents for real-world applications
• 2. Building code-first agents with Strands SDK
• 3. Integrating agents with external APIs and dynamic tool discovery through Model Context Protocol
• 4. Implementing persistent memory and conversation state management
• 5. Applying RAG patterns for private data access
• 6. Connecting foundation models with external tools and services
• 7. Exposing agents through secure API endpoints with authentication

Practice Exam 1 - AWS Certified Generative AI Developer – Professional