AWS Certified Generative AI Developer — Professional (AIP-C01)

A documentation-first study guide. AWS writes the exam from its own documentation, so reading the docs is the highest-leverage thing you can do. This guide is a curated index into the canonical references, FAQs, and a selection of whitepapers — organised around the five exam domains and their official task statements, not around services.

Maps to the published AWS Certified Generative AI Developer — Professional (AIP-C01) exam guide. Domain weights and the numbered skill statements inside each requirement block are quoted verbatim from that PDF (Exam Guide version 1.0). Last verified against the official exam guide and AWS documentation on 2026-07-08.

About the exam

Current exam code: AIP-C01. This is AWS's first professional-level generative AI certification. It ran as a beta from November 2025 through March 2026; the standard (non-beta) exam has been generally available since mid-2026.

Format: 75 questions (65 scored + 10 unscored) · 180 minutes · $300 USD · scaled score 100–1000, pass at 750. The 65-scored count applies to the standard exam; the beta form was longer.

Question types: Multiple choice (one correct response, three distractors), multiple response (two or more correct responses out of five or more options — select all to get credit), ordering (place 3–5 responses in the correct order), and matching (match a list of responses to 3–7 prompts). You must get every part correct to receive credit on ordering and matching questions.

The five domains:

Domain 1 — Foundation Model Integration, Data Management, and Compliance — 31%
Domain 2 — Implementation and Integration — 26%
Domain 3 — AI Safety, Security, and Governance — 20%
Domain 4 — Operational Efficiency and Optimization for GenAI Applications — 12%
Domain 5 — Testing, Validation, and Troubleshooting — 11%

Primary official sources (bookmark these):

Key resources:

Amazon Bedrock User Guide — the primary service for this exam
Exam Prep Plan: AWS Certified Generative AI Developer — Professional (Skill Builder) — official exam-prep learning plan
AWS Well-Architected Framework — Machine Learning Lens
AWS Well-Architected Tool — the exam guide references the Generative AI Lens and the WA Tool for standardised solution design

Priority tiers: The published domain weights (31/26/20/12/11) tell you how the exam is balanced across the five domains, but they don't tell you that within each domain a handful of services account for most of the questions. Every section in this guide carries a tier badge based on triangulating the AWS exam guide, the experience reports of recent test takers, and the patterns that appear in the practice-exam community:

★★★ Core Heavily tested. Multiple questions will lean on this. Spend hours, not minutes — if you don't know it well, you fail.
★★ Important Reliably tested, usually one or two questions. Read every linked page in the section, do the FAQ, understand the comparison points. A few hours per topic.
★ Light Known to appear, but typically as one distinguishing question or as wrong-answer distractors. Skim the docs, learn the one-line distinction, move on. Twenty minutes to an hour.

For an 8–12 week prep cycle the rough split that the data supports is about 60% of your time on Core topics, 30% on Important, and 10% on Light. The biggest single concentration of questions across the whole exam is the cluster around Amazon Bedrock plus Knowledge Bases, prompt engineering, RAG patterns, and agents — know those cold and you have the foundation of a pass.

How to use this guide:

Each section opens with a one-paragraph summary explaining what to focus on, then has up to three link sections: Core docs (user/developer guides — the canonical reference), FAQ (exam writers love edge cases from FAQs — do not skip), and Deeper reading (whitepapers, blog posts, re:Post articles).
If a link 404s, AWS has reorganised the docs. Search the page title to find the new location — the content almost always still exists.
The What's New feed is worth a weekly scan in the last month before your exam; generative AI is fast-moving but the exam lags new features by roughly 6 to 12 months.

Part I — Domain 1: Foundation Model Integration, Data Management, and Compliance (31%)

The largest domain by weight. It covers solution design, selecting and configuring foundation models, data pipelines for FM consumption, vector stores, retrieval mechanisms for augmentation (RAG), and prompt engineering and governance. Six official tasks live here.

Chapter 1 — Solution design and foundation models

Maps to Task Statement 1.1 — Analyze requirements and design GenAI solutions

Skill 1.1.1 — Create comprehensive architectural designs that align with specific business needs and technical constraints (for example, by using appropriate FMs, integration patterns, deployment strategies).
Skill 1.1.2 — Develop technical proof-of-concept implementations to validate feasibility, performance characteristics, and business value before proceeding to full-scale deployment (for example, by using Amazon Bedrock).
Skill 1.1.3 — Create standardized technical components to ensure consistent implementation across multiple deployment scenarios (for example, by using the AWS Well-Architected Framework, AWS WA Tool Generative AI Lens).

1.1 Foundation model concepts ★★★ Core

Understand what foundation models are, how they differ from traditional ML models, and the key concepts: pre-training, fine-tuning, inference, tokens, context windows, temperature, top-p, and stop sequences. Know the trade-offs between model size, latency, and cost. This conceptual foundation underpins everything else.

Core docs

What is a foundation model? — pre-trained on massive datasets, adaptable to many tasks
Generative AI on AWS overview — service landscape and use cases
Models at a glance — available models in Bedrock, capabilities, and limits
Inference parameters — temperature, top-p, max tokens, stop sequences

Deeper reading

Exam Prep Plan: Generative AI Developer — Professional (Skill Builder) — official free learning plan covering FM concepts and AWS services

1.2 Amazon Bedrock overview ★★★ Core

Amazon Bedrock is the central service for this exam. Know what Bedrock is (fully managed service for accessing foundation models via API), the available model providers (Amazon Nova, Amazon Titan, Anthropic Claude, Meta Llama, Mistral, Cohere, AI21 Labs, Stability AI), and how to access models. Understand model access requests, pricing models (on-demand vs provisioned throughput), and regional availability including cross-region inference.

Core docs

What is Amazon Bedrock? — fully managed service for accessing FMs via API
Supported foundation models — all available models, providers, and capabilities
Model access — requesting access to specific models
Amazon Bedrock pricing — on-demand vs provisioned, per-token costs
Provisioned throughput — reserved capacity for consistent performance
Cross-region inference — automatic routing to available regions

FAQ

Amazon Bedrock FAQ — common questions on data privacy, model access, pricing

Maps to Task Statement 1.2 — Select and configure FMs

Skill 1.2.1 — Assess and choose FMs to ensure optimal alignment with specific business use cases and technical requirements (for example, by using performance benchmarks, capability analysis, limitation evaluation).
Skill 1.2.2 — Create flexible architecture patterns to enable dynamic model selection and provider switching without requiring code modifications (for example, by using AWS Lambda, Amazon API Gateway, AWS AppConfig).
Skill 1.2.3 — Design resilient AI systems to ensure continuous operation during service disruptions (for example, by using AWS Step Functions circuit breaker patterns, Amazon Bedrock Cross-Region Inference for models that have limited regional availability, cross-Region model deployment, graceful degradation strategies).
Skill 1.2.4 — Implement FM customization deployment and lifecycle management (for example, by using Amazon SageMaker AI to deploy domain-specific fine-tuned models, parameter-efficient adaptation techniques such as low-rank adaptation [LoRA] and adapters for model deployment, SageMaker Model Registry for versioning and to deploy customized models, automated deployment pipelines to update models, rollback strategies for failed deployments, lifecycle management to retire and replace models).

1.3 Amazon Nova models ★★★ Core

Amazon Nova is AWS's current flagship foundation model family and a likely exam focus. Know the tiers: Nova Micro (text-only, lowest latency and cost), Nova Lite and Nova Pro (multimodal — text, image, video input), Nova Premier (most capable for complex reasoning), plus Nova Canvas (image generation), Nova Reel (video generation), and Nova Sonic (speech-to-speech). Understand when to pick a smaller Nova tier for cost and latency versus a larger tier for quality.

Core docs

Amazon Nova models — capabilities and parameters for the flagship multimodal tier
Supported foundation models — the full Nova line-up alongside third-party providers
Amazon Titan multimodal embeddings — joint text-image embeddings (still current for RAG, see Chapter 3)
Inference parameters — tune temperature, top-p, and max tokens per Nova tier

1.4 Third-party models in Bedrock ★★ Important

Know the key third-party models available: Anthropic Claude (advanced reasoning, long context, vision), Meta Llama (open-weight models), Mistral (efficient inference, multilingual), Cohere (Command for generation, Embed for embeddings, Rerank for reranking), AI21 Labs Jamba (long-context hybrid SSM-Transformer models — Jurassic is retired), and Stability AI (image generation). Model selection is Skill 1.2.1: match capability, latency, cost, and regional availability to the use case.

Core docs

Anthropic Claude models — advanced reasoning, long context, vision capabilities
Meta Llama models — open-weight models, cost-effective inference
Mistral AI models — efficient inference, multilingual support
Cohere models — Command for generation, Embed for embeddings
AI21 Labs models — Jamba family for long-context enterprise text
Stability AI models — image generation and editing

1.5 Model customization: fine-tuning and PEFT ★★ Important

Know when and how to customize models (Skill 1.2.4). Understand the difference between continued pre-training (domain adaptation on unlabelled data) and fine-tuning (task-specific training on labelled examples), and parameter-efficient techniques such as low-rank adaptation (LoRA) and adapters. Know data preparation requirements and how customized models are deployed and versioned (for example, with the SageMaker AI Model Registry).

Core docs

Custom models in Bedrock — overview of customization options, including continued pre-training and fine-tuning
Fine-tuning a model — task-specific training with labelled examples
Prepare the datasets — format requirements and data-quality guidelines
Submit a customization job — job configuration and hyperparameters
SageMaker Model Registry — version and deploy customized models (Skill 1.2.4)

Chapter 2 — Data pipelines for FM consumption

Maps to Task Statement 1.3 — Implement data validation and processing pipelines for FM consumption

Skill 1.3.1 — Create comprehensive data validation workflows to ensure data meets quality standards for FM consumption (for example, by using AWS Glue Data Quality, SageMaker Data Wrangler, custom Lambda functions, Amazon CloudWatch metrics).
Skill 1.3.2 — Create data processing workflows to handle complex data types including text, image, audio, and tabular data with specialized processing requirements for FM consumption (for example, by using Amazon Bedrock multimodal models, SageMaker Processing, AWS Transcribe, advanced multimodal pipeline architectures).
Skill 1.3.3 — Format input data for FM inference according to model-specific requirements (for example, by using JSON formatting for Amazon Bedrock API requests, structured data preparation for SageMaker AI endpoints, conversation formatting for dialog-based applications).
Skill 1.3.4 — Enhance input data quality to improve FM response quality and consistency (for example, by using Amazon Bedrock to reformat text, Amazon Comprehend to extract entities, Lambda functions to normalize data).

2.1 Data validation and processing ★★ Important

Know how to prepare data for RAG and fine-tuning. Understand data validation (AWS Glue Data Quality, SageMaker Data Wrangler, CloudWatch metrics), document parsing, handling multimodal data (text, image, audio, tabular), and formatting input for FM inference (JSON for Bedrock API requests, conversation formatting for dialog apps).

Core docs

AWS Glue Data Quality — validate data quality before FM consumption
Amazon SageMaker Data Wrangler — visual data preparation and validation
Amazon Comprehend entity recognition — extract entities to enrich input data
Amazon Transcribe — convert audio to text for multimodal pipelines

2.2 Amazon Bedrock Data Automation ★★ Important

Know Amazon Bedrock Data Automation (BDA) for turning unstructured multimodal content — documents, images, audio, video — into structured output usable by FMs and knowledge bases. Understand blueprints for custom output, and how BDA fits document processing and RAG ingestion pipelines (Skills 1.3.2, 2.5.3).

Core docs

Amazon Bedrock Data Automation — extract structured data from documents, images, audio, and video
How Bedrock Data Automation works — standard output and blueprints

Chapter 3 — Vector stores, embeddings, and retrieval (RAG)

RAG and vector stores are officially part of Domain 1 (Tasks 1.4 and 1.5), not Domain 2. This is the single densest cluster of questions on the exam.

Maps to Task Statement 1.4 — Design and implement vector store solutions

Skill 1.4.1 — Create advanced vector database architectures specifically for FM augmentation to enable efficient semantic retrieval beyond traditional search capabilities (for example, by using Amazon Bedrock Knowledge Bases for hierarchical organization, Amazon OpenSearch Service with the Neural plugin for Amazon Bedrock integration for topic-based segmentation, Amazon RDS with Amazon S3 document repositories, Amazon DynamoDB with vector databases for metadata and embeddings).
Skill 1.4.2 — Develop comprehensive metadata frameworks to improve search precision and context awareness for FM interactions (for example, by using S3 object metadata for document timestamps, custom attributes for authorship information, tagging systems for domain classification).
Skill 1.4.3 — Implement high-performance vector database architectures to optimize semantic search performance at scale for FM retrieval (for example, by using OpenSearch sharding strategies, multi-index approaches for specialized domains, hierarchical indexing techniques).
Skill 1.4.4 — Use AWS services to create integration components to connect with resources (for example, document management systems, knowledge bases, internal wikis for comprehensive data integration in GenAI applications).
Skill 1.4.5 — Design and deploy data maintenance systems to ensure that vector stores contain current and accurate information for FM augmentation (for example, by using incremental update mechanisms, real-time change detection systems, automated synchronization workflows, scheduled refresh pipelines).

Maps to Task Statement 1.5 — Design retrieval mechanisms for FM augmentation

Skill 1.5.1 — Develop effective document segmentation approaches to optimize retrieval performance for FM context augmentation (for example, by using Amazon Bedrock chunking capabilities, Lambda functions to implement fixed-size chunking, custom processing for hierarchical chunking based on content structure).
Skill 1.5.2 — Select and configure optimal embedding solutions to create efficient vector representations for semantic search (for example, by using Amazon Titan embeddings based on dimensionality and domain fit, by evaluating performance characteristics of Amazon Bedrock embedding models, by using Lambda functions to batch generate embeddings).
Skill 1.5.3 — Deploy and configure vector search solutions to enable semantic search capabilities for FM augmentation (for example, by using OpenSearch Service with vector search capabilities, Amazon Aurora with the pgvector extension, Amazon Bedrock Knowledge Bases with managed vector store functionality).
Skill 1.5.4 — Create advanced search architectures to improve the relevance and accuracy of retrieved information for FM context (for example, by using OpenSearch for semantic search, hybrid search that combines keywords and vectors, Amazon Bedrock reranker models).
Skill 1.5.5 — Develop sophisticated query handling systems to improve the retrieval effectiveness and result quality for FM augmentation (for example, by using Amazon Bedrock for query expansion, Lambda functions for query decomposition, Step Functions for query transformation).
Skill 1.5.6 — Create consistent access mechanisms to enable seamless integration with FMs (for example, by using function calling interfaces for vector search, Model Context Protocol [MCP] clients for vector queries, standardized API patterns for retrieval augmentation).

3.1 Vector embeddings ★★★ Core

Vector embeddings are fundamental to RAG. Know what embeddings are (dense vector representations of text or images), how to generate them (Amazon Titan Text Embeddings, Cohere Embed), and their role in semantic search. Understand embedding dimensions, normalization, and similarity metrics (cosine, dot product). Titan Text Embeddings remains the default embedding choice on Bedrock even as Nova leads generation.

Core docs

Amazon Titan Text Embeddings — dense vectors for semantic search
Titan Text Embeddings V2 — configurable dimensions, normalisation options
Cohere Embed models — multilingual embeddings, input types
Titan Multimodal Embeddings — joint text-image embedding space

3.2 RAG concepts ★★★ Core

RAG is heavily tested. Know the RAG pattern (retrieve relevant context, augment the prompt, generate the response), why it's used (reduce hallucinations, incorporate proprietary data), and the components involved (embeddings, vector store, retrieval, generation). Understand chunking strategies (fixed-size, semantic, hierarchical) and the trade-offs between retrieval quality and latency.

Core docs

Retrieval Augmented Generation with Knowledge Bases — retrieve, augment, and generate pattern
How knowledge bases work — embedding, indexing, retrieval, generation flow
How retrieval works — chunking, metadata, and retrieval tuning

3.3 Bedrock Knowledge Bases ★★★ Core

Bedrock Knowledge Bases is the managed RAG service. Know how to create knowledge bases, configure data sources (S3, web crawlers, Confluence, SharePoint), set up vector stores (OpenSearch Serverless, Aurora, Amazon Neptune Analytics, Pinecone, Redis), and query knowledge bases. Understand sync jobs and incremental updates (Skill 1.4.5).

Core docs

Bedrock Knowledge Bases — managed RAG service overview
Creating a knowledge base — step-by-step setup guide
Connect a data source — supported sources and configuration
Vector store prerequisites — OpenSearch, Aurora, Pinecone, Redis setup
Query and retrieve data — testing retrieval in the console
RetrieveAndGenerate API — single API call for RAG

3.4 Vector databases ★★ Important

Know the vector database options for Bedrock Knowledge Bases: Amazon OpenSearch Serverless (fully managed, scalable), Aurora PostgreSQL with pgvector, Amazon Neptune Analytics, Pinecone, and Redis. Understand selection criteria (scale, latency, cost, features), plus OpenSearch sharding and multi-index strategies (Skill 1.4.3).

Core docs

Vector store prerequisites — supported vector databases and setup
OpenSearch Serverless — fully managed, scales automatically
Aurora with pgvector — PostgreSQL extension for vector search
OpenSearch k-NN — similarity search algorithms and index types

3.5 Retrieval optimization and reranking ★★ Important

Know how to raise retrieval relevance beyond a single vector search (Skills 1.5.4 to 1.5.6): hybrid search that combines keyword and vector matching, Amazon Bedrock reranker models to reorder candidate results, and query handling (expansion, decomposition, transformation). Know that MCP clients and function-calling interfaces provide consistent access to retrieval for FMs.

Core docs

Rerank in Amazon Bedrock — reorder retrieved results to improve relevance
Use a reranker model — apply reranking in retrieval pipelines
How retrieval works — hybrid search and query configuration

Chapter 4 — Prompt engineering and management

Maps to Task Statement 1.6 — Implement prompt engineering strategies and governance for FM interactions

Skill 1.6.1 — Create effective model instruction frameworks to control FM behavior and outputs (for example, by using Amazon Bedrock Prompt Management to enforce role definitions, Amazon Bedrock Guardrails to enforce responsible AI guidelines, template configurations to format responses).
Skill 1.6.2 — Build interactive AI systems to maintain context and improve user interactions with FMs (for example, by using Step Functions for clarification workflows, Amazon Comprehend for intent recognition, DynamoDB for conversation history storage).
Skill 1.6.3 — Implement comprehensive prompt management and governance systems to ensure consistency and oversight of FM operations (for example, by using Amazon Bedrock Prompt Management to create parameterized templates and approval workflows, Amazon S3 to store template repositories, AWS CloudTrail to track usage, Amazon CloudWatch Logs to log access).
Skill 1.6.4 — Develop quality assurance systems to ensure prompt effectiveness and reliability for FMs (for example, by using Lambda functions to verify expected output, Step Functions to test edge cases, CloudWatch to test prompt regression).
Skill 1.6.5 — Enhance FM performance to refine prompts iteratively and improve response quality beyond basic prompting techniques (for example, by using structured input components, output format specifications, chain-of-thought instruction patterns, feedback loops).
Skill 1.6.6 — Design complex prompt systems to handle sophisticated tasks with FMs (for example, by using Amazon Bedrock Prompt Flows for sequential prompt chains, conditional branching based on model responses, reusable prompt components, integrated pre-processing and post-processing steps).

4.1 Prompt engineering fundamentals ★★★ Core

Prompt engineering is heavily tested. Know the components of effective prompts (system prompts, user prompts, context, examples), prompt patterns (zero-shot, few-shot, chain-of-thought), and techniques for improving output quality. Understand how to structure prompts for different tasks (summarization, extraction, generation, classification).

Core docs

Prompt engineering concepts — best practices for effective prompts
Design a prompt — step-by-step guide to prompt construction
Prompt templates and examples — reusable patterns for common tasks
What is prompt engineering? — components: system prompts, user prompts, context

Deeper reading

Prompt engineering overview (Anthropic) — zero-shot, few-shot, chain-of-thought patterns

4.2 Bedrock Prompt Management ★★ Important

Know how to use Bedrock Prompt Management to create, version, and govern prompts. Understand prompt templates with variables (placeholders), prompt testing, and versioning and approval workflows for production (Skills 1.6.1, 1.6.3).

Core docs

Bedrock Prompt Management — store, version, and manage prompts centrally
Creating prompts — console and API methods
Prompt placeholders and variables — dynamic content in templates
Prompt versions — track changes and roll back

4.3 Bedrock Flows ★★ Important

Know Bedrock Flows (formerly Prompt Flows; the exam guide still uses the name "Prompt Flows") for orchestrating multi-step generative AI workflows. Understand nodes (prompt, condition, iterator, collector, Lambda, knowledge base, agent), flow connections, and how to chain prompts for complex tasks (Skill 1.6.6).

Core docs

Bedrock Flows — visual orchestration of multi-step AI workflows
Create a flow — building flows in the console or with the API
Flow nodes — prompt, condition, iterator, collector, Lambda nodes
Test a flow — validate flow behaviour before deployment

Part II — Domain 2: Implementation and Integration (26%)

Building production generative AI applications. This covers agentic AI, model deployment strategies, enterprise integration architectures, FM API integrations, and application integration patterns and developer tools. Five official tasks live here.

Chapter 5 — Agentic AI

Maps to Task Statement 2.1 — Implement agentic AI solutions and tool integrations

Skill 2.1.1 — Develop intelligent autonomous systems with appropriate memory and state management capabilities (for example, by using Strands Agents and AWS Agent Squad for multi-agent systems, MCP for agent-tool interactions).
Skill 2.1.2 — Create advanced problem-solving systems to give FMs the ability to break down and solve complex problems by following structured reasoning steps (for example, by using Step Functions to implement ReAct patterns and chain-of-thought reasoning approaches).
Skill 2.1.3 — Develop safeguarded AI workflows to ensure controlled FM behavior (for example, by using Step Functions to implement stopping conditions, Lambda functions to implement timeout mechanisms, IAM policies to enforce resource boundaries, circuit breakers to mitigate failures).
Skill 2.1.4 — Create sophisticated model coordination systems to optimize performance across multiple capabilities (for example, by using specialized FMs to perform complex tasks, custom aggregation logic for model ensembles, model selection frameworks).
Skill 2.1.5 — Develop collaborative AI systems to enhance FM capabilities with human expertise (for example, by using Step Functions to orchestrate review and approval processes, API Gateway to implement feedback collection mechanisms, human augmentation patterns).
Skill 2.1.6 — Implement intelligent tool integrations to extend FM capabilities and to ensure reliable tool operations (for example, by using the Strands API to implement custom behaviors, standardized function definitions, Lambda functions to implement error handling and parameter validation).
Skill 2.1.7 — Develop model extension frameworks to enhance FM capabilities (for example, by using Lambda functions to implement stateless MCP servers that provide lightweight tool access, Amazon ECS to implement MCP servers that provide complex tools, MCP client libraries to ensure consistent access patterns).

5.1 Agents for Amazon Bedrock ★★★ Core

Agents orchestrate multi-step tasks by using foundation models to reason, plan, and take actions. Know how to create agents, define action groups (Lambda functions, API schemas), connect knowledge bases, and configure agent instructions. Understand the agent execution flow and how agents break down complex tasks.

Core docs

Agents for Amazon Bedrock — orchestrate multi-step tasks with FMs
How agents work — reasoning, planning, and action execution flow
Creating an agent — configuration, instructions, model selection
Create an action group — connect agents to external APIs and Lambda
Advanced prompts and agent instructions — guide agent behaviour
Add a knowledge base to an agent — RAG-enabled agents for grounded responses

5.2 Agent action groups and APIs ★★ Important

Know how to define action groups using OpenAPI schemas or Lambda functions. Understand how agents invoke actions, pass parameters, and handle responses. Know about return of control for human-in-the-loop patterns and how agents are versioned and deployed with aliases.

Core docs

Action group Lambda functions — Lambda as action executor
OpenAPI schemas for action groups — define APIs agents can call
Return of control — human-in-the-loop patterns
Deploy an agent with aliases — deployment stages and version management

5.3 Multi-agent systems: Strands Agents and AWS Agent Squad ★★ Important

The exam guide names Strands Agents and AWS Agent Squad explicitly (Skills 2.1.1, 2.1.6, 2.5.5). Strands Agents is an open-source, model-driven agent SDK; AWS Agent Squad (formerly Multi-Agent Orchestrator) coordinates multiple specialized agents. Know when to use AWS-native multi-agent orchestration versus a single Bedrock agent, and how these frameworks manage memory and tool use.

Core docs

Multi-agent collaboration in Bedrock — a supervisor agent coordinating specialized agents
Create a multi-agent collaboration — configure collaborator agents
Introducing Strands Agents (AWS Open Source Blog) — open-source, model-driven agent framework

5.4 Model Context Protocol (MCP) ★★ Important

MCP is called out repeatedly (Skills 2.1.1, 2.1.7, 1.5.6). It is an open standard for connecting FMs and agents to tools and data through MCP servers and clients. Know that stateless MCP servers can run on Lambda for lightweight tool access and on Amazon ECS for heavier tools, and that MCP client libraries give agents consistent access patterns to vector search and other resources.

Core docs

Model Context Protocol introduction — the open MCP standard for tool and data access
Amazon Bedrock AgentCore Gateway — turn APIs and Lambda functions into MCP-compatible tools for agents

5.5 Amazon Bedrock AgentCore ★ Light

AgentCore is a separate service (its own documentation set, not part of the Bedrock user guide) for deploying and operating agents securely at scale, regardless of framework or model. Know its building blocks — Runtime, Memory, Gateway, Identity, and Observability — and that it complements, rather than replaces, managed Bedrock Agents.

Core docs

What is Amazon Bedrock AgentCore? — enterprise-grade infrastructure to deploy and operate agents
AgentCore Gateway — connect agents to tools via MCP
AgentCore Memory — short-term and long-term memory for agents

Chapter 6 — Model deployment strategies

Maps to Task Statement 2.2 — Implement model deployment strategies

Skill 2.2.1 — Deploy FMs based on specific application needs and performance requirements (for example, by using Lambda functions for on-demand invocation, Amazon Bedrock provisioned throughput configurations, SageMaker AI endpoints to implement hybrid solutions).
Skill 2.2.2 — Deploy FM solutions by addressing unique challenges of large language models (LLMs) that differ from traditional ML deployments (for example, by implementing container-based deployment patterns that are optimized for memory requirements, GPU utilization, and token processing capacity, by following specialized model loading strategies).
Skill 2.2.3 — Develop optimized FM deployment approaches to balance performance and resource requirements for GenAI workloads (for example, by selecting appropriate models, by using smaller pre-trained models for specific tasks, by using API-based model cascading to perform routine queries).

6.1 Bedrock, SageMaker, and Lambda deployment options ★★ Important

Know the deployment spectrum: Lambda for on-demand serverless invocation, Bedrock provisioned throughput for predictable capacity, and SageMaker AI endpoints for hosting custom or open-weight models (GPU-optimized, container-based). Understand model cascading — routing routine queries to smaller, cheaper models and escalating only when needed (Skill 2.2.3).

Core docs

Provisioned throughput — reserved capacity for consistent performance
SageMaker JumpStart — deploy pre-trained FMs to endpoints
SageMaker real-time inference — host custom models for low-latency inference
Invoke a model with the Bedrock API — on-demand invocation from Lambda and other compute

Chapter 7 — Enterprise integration architectures

Maps to Task Statement 2.3 — Design and implement enterprise integration architectures

Skill 2.3.1 — Create enterprise connectivity solutions to seamlessly incorporate FM capabilities into existing enterprise environments (for example, by using API-based integrations with legacy systems, event-driven architectures to implement loose coupling, data synchronization patterns).
Skill 2.3.2 — Develop integrated AI capabilities to enhance existing applications with GenAI functionality (for example, by using API Gateway to implement microservice integrations, Lambda functions for webhook handlers, Amazon EventBridge to implement event-driven integrations).
Skill 2.3.3 — Create secure access frameworks to ensure appropriate security controls (for example, by using identity federation between FM services and enterprise systems, role-based access control for model and data access, least privilege API access to FMs).
Skill 2.3.4 — Develop cross-environment AI solutions to ensure data compliance across jurisdictions while enabling FM access (for example, by using AWS Outposts for on-premises data integration, AWS Wavelength to perform edge deployments, secure routing between cloud and on-premises resources).
Skill 2.3.5 — Implement CI/CD pipelines and GenAI gateway architectures to implement secure and compliant consumption patterns in enterprise environments (for example, by using AWS CodePipeline, AWS CodeBuild, automated testing frameworks for continuous deployment and testing of GenAI components with security scans and rollback support, centralized abstraction layers, observability and control mechanisms).

7.1 Event-driven and enterprise connectivity ★★ Important

Know how to weave FMs into existing systems with loose coupling: API Gateway for microservice and webhook integration, EventBridge for event-driven triggers, and data synchronization patterns. Understand cross-jurisdiction options — AWS Outposts for on-premises data and AWS Wavelength for edge — when data residency constrains where inference can run (Skill 2.3.4).

Core docs

Amazon EventBridge — event-driven integration across services
Amazon API Gateway — front FMs with managed, secured APIs
AWS Outposts — on-premises AWS for data-residency-constrained integration
AWS Wavelength — edge deployment for low-latency access

7.2 CI/CD and GenAI gateways ★ Light

Know how to build secure, compliant consumption patterns: CI/CD pipelines (CodePipeline, CodeBuild) with automated testing, security scans, and rollback for GenAI components, plus a centralized GenAI gateway that provides an abstraction layer, observability, and cost and access control across teams (Skill 2.3.5).

Core docs

AWS CodePipeline — continuous delivery orchestration
AWS CodeBuild — build, test, and security-scan artifacts

Chapter 8 — FM API integrations

Maps to Task Statement 2.4 — Implement FM API integrations

Skill 2.4.1 — Create flexible model interaction systems (for example, by using Amazon Bedrock APIs to manage synchronous requests from various compute environments, language-specific AWS SDKs and Amazon SQS for asynchronous processing, API Gateway to provide custom API clients with request validation).
Skill 2.4.2 — Develop real-time AI interaction systems to provide immediate feedback from FM (for example, by using Amazon Bedrock streaming APIs for incremental response delivery, WebSockets or server-sent events to generate text in real time, API Gateway to implement chunked transfer encoding).
Skill 2.4.3 — Create resilient FM systems to ensure reliable operations (for example, by using the AWS SDK for exponential backoff, API Gateway to manage rate limiting, fallback mechanisms for graceful degradation, AWS X-Ray to provide observability across service boundaries).
Skill 2.4.4 — Develop intelligent model routing systems to optimize model selection (for example, by using application code to implement static routing configurations, Step Functions for dynamic content-based routing to specialized FMs, intelligent model routing based on metrics, API Gateway with request transformations for routing logic).

8.1 Bedrock APIs and SDKs ★★★ Core

Know how to invoke Bedrock models programmatically. Understand the InvokeModel and Converse APIs, request and response formats for different models, and how to use the AWS SDKs (Python Boto3, JavaScript). Know resilient-client patterns: exponential backoff, rate-limit handling, and fallbacks (Skill 2.4.3).

Core docs

Bedrock API Reference — complete API documentation
InvokeModel API — synchronous model invocation
Converse API — unified API for multi-turn conversations
Using Bedrock with Boto3 — Python SDK reference

8.2 Streaming and real-time applications ★★ Important

Know how to implement streaming responses for better user experience (Skill 2.4.2). Understand the InvokeModelWithResponseStream API, Server-Sent Events (SSE), WebSocket patterns with API Gateway, and how to handle partial responses in the UI.

Core docs

InvokeModelWithResponseStream API — token-by-token output for lower latency
WebSocket APIs in API Gateway — bidirectional real-time communication
Lambda response streaming — stream large responses from Lambda

Chapter 9 — Application integration patterns and developer tools

Maps to Task Statement 2.5 — Implement application integration patterns and development tools

Skill 2.5.1 — Create FM API interfaces to address the specific requirements of GenAI workloads (for example, by using API Gateway to handle streaming responses, token limit management, retry strategies to handle model timeouts).
Skill 2.5.2 — Develop accessible AI interfaces to accelerate adoption and integration of FMs (for example, by using AWS Amplify to develop declarative UI components, OpenAPI specifications for API-first development approaches, Amazon Bedrock Prompt Flows for no-code workflow builders).
Skill 2.5.3 — Create business system enhancements (for example, by using Lambda functions to implement customer relationship management [CRM] enhancements, Step Functions to orchestrate document processing systems, Amazon Q Business data sources to provide internal knowledge tools, Amazon Bedrock Data Automation to manage automated data processing workflows).
Skill 2.5.4 — Enhance developer productivity to accelerate development workflows for GenAI applications (for example, by using Amazon Q Developer to generate and refactor code, code suggestions for API assistance, AI component testing, performance optimization).
Skill 2.5.5 — Develop advanced GenAI applications to implement sophisticated AI capabilities (for example, by using Strands Agents and AWS Agent Squad for AWS native orchestration, Step Functions to orchestrate agent design patterns, Amazon Bedrock to manage prompt chaining patterns).
Skill 2.5.6 — Improve troubleshooting efficiency for FM applications (for example, by using CloudWatch Logs Insights to analyze prompts and responses, X-Ray to trace FM API calls, Amazon Q Developer to implement GenAI-specific error pattern recognition).

9.1 AWS Amplify and API interfaces ★ Light

Know how to accelerate FM adoption with accessible interfaces (Skill 2.5.2): the AWS Amplify AI kit for declarative front-end components backed by Bedrock, OpenAPI specifications for API-first design, and AWS AppSync GraphQL APIs with resolvers that call Bedrock.

Core docs

AWS Amplify AI kit — integrate Bedrock into web and mobile apps
AWS AppSync — managed GraphQL service
Configuring resolvers — connect GraphQL operations to data sources
JavaScript resolvers reference — custom resolver logic

9.2 Amazon Q Business and Q Developer ★★ Important

Know Amazon Q Business for enterprise AI assistants (data connectors, guardrails, user context) as an internal-knowledge tool (Skill 2.5.3), and Amazon Q Developer for coding assistance — generation, refactoring, code reviews, and GenAI-specific error pattern recognition (Skills 2.5.4, 2.5.6). Understand how Q Business differs from building custom RAG with Bedrock Knowledge Bases.

Core docs

What is Amazon Q Business? — enterprise AI assistant with built-in RAG
Setting up Amazon Q Business — application and index configuration
Data connectors — enterprise sources (SharePoint, Salesforce, and more)
What is Amazon Q Developer? — AI coding assistant for IDEs and the CLI
Code reviews with Q Developer — identify issues and vulnerabilities in code

FAQ

Amazon Q Business FAQ — licensing, data handling, supported sources

9.3 Amazon Lex conversational interfaces ★★ Important

Know Amazon Lex for building conversational interfaces. Understand intents, slots, fulfillment with Lambda, and Lex's generative features that integrate Bedrock for richer responses (for example, QnAIntent). Know when to use Lex versus direct Bedrock integration, and how Amazon Comprehend supports intent recognition (Skill 1.6.2).

Core docs

What is Amazon Lex? — conversational interfaces with voice and text
Building bots — bot design and configuration
Intents and slots — user intent recognition, parameter extraction
Generative AI features in Lex — enhance Lex with Bedrock foundation models

FAQ

Amazon Lex FAQ — pricing, languages, integration options

9.4 Amazon Connect with GenAI ★ Light

Know how Amazon Connect integrates with generative AI for contact center applications: Amazon Q in Connect for real-time agent assistance, and Lex integration for conversational IVR.

Core docs

Amazon Connect — cloud contact centre service
Amazon Q in Connect setup — real-time agent assistance with GenAI
Amazon Connect with Lex — conversational IVR integration

Part III — Domain 3: AI Safety, Security, and Governance (20%)

Safety controls, data security and privacy, governance and compliance, and responsible AI for generative AI applications. Four official tasks live here.

Chapter 10 — Input and output safety controls

Maps to Task Statement 3.1 — Implement input and output safety controls

Skill 3.1.1 — Develop comprehensive content safety systems to protect against harmful user inputs to FMs (for example, by using Amazon Bedrock guardrails to filter content, Step Functions and Lambda functions to implement custom moderation workflows, real-time validation mechanisms).
Skill 3.1.2 — Create content safety frameworks to prevent harmful outputs (for example, by using Amazon Bedrock guardrails to filter responses, specialized FM evaluations for content moderation and toxicity detection, text-to-SQL transformations to ensure deterministic results).
Skill 3.1.3 — Develop accuracy verification systems to reduce hallucinations in FM responses (for example, by using Amazon Bedrock Knowledge Base to ground responses and perform fact-checking, confidence scoring and semantic similarity search for verification, JSON Schema to enforce structured outputs).
Skill 3.1.4 — Create defense-in-depth safety systems to provide comprehensive protection against FM misuse (for example, by using Amazon Comprehend to develop pre-processing filters, Amazon Bedrock to implement model-based guardrails, Lambda functions to perform post-processing validation, API Gateway to implement API response filtering).
Skill 3.1.5 — Implement advanced threat detection to protect against adversarial inputs and security vulnerabilities (for example, by using prompt injection and jailbreak detection mechanisms, input sanitization and content filters, safety classifiers, automated adversarial testing workflows).

10.1 Bedrock Guardrails ★★★ Core

Guardrails is heavily tested. Know how to configure guardrails for content filtering (hate, insults, sexual, violence, misconduct, prompt attacks), sensitive-information (PII) detection and redaction, denied topics, and word filters. Understand contextual grounding checks to reduce hallucinations, and how guardrails apply to both inputs and outputs.

Core docs

Guardrails for Amazon Bedrock — safety controls for inputs and outputs
Components of a guardrail — filters, topics, and policies you configure
Content filters — hate, insults, sexual, violence, misconduct, prompt attacks
Denied topics — block specific conversation topics
Sensitive information filters — detect and redact PII
Word filters — block specific words and phrases
Contextual grounding check — reduce hallucinations against source context
Testing guardrails — validate filter effectiveness

10.2 Threat detection and hallucination reduction ★★ Important

Know defense-in-depth beyond guardrails (Skills 3.1.3 to 3.1.5): prompt injection and jailbreak detection, input sanitization, safety classifiers, grounding responses in a knowledge base for fact-checking, and JSON Schema to enforce structured, deterministic outputs.

Core docs

Prompt attack protection — detect prompt injection and jailbreak attempts
Ground responses with a knowledge base — reduce hallucinations with retrieval
Amazon Comprehend PII detection — pre-processing filters for sensitive input

Chapter 11 — Data security and privacy controls

Maps to Task Statement 3.2 — Implement data security and privacy controls

Skill 3.2.1 — Develop protected AI environments to ensure comprehensive security for FM deployments (for example, by using VPC endpoints to isolate networks, IAM policies to enforce secure data access patterns, AWS Lake Formation to provide granular data access, CloudWatch to monitor data access).
Skill 3.2.2 — Develop privacy-preserving systems to protect sensitive information during FM interactions (for example, by using Amazon Comprehend and Amazon Macie to detect personally identifiable information [PII], Amazon Bedrock native data privacy features, Amazon Bedrock guardrails to filter outputs, Amazon S3 Lifecycle configurations to implement data retention policies).
Skill 3.2.3 — Create privacy-focused AI systems to protect user privacy while maintaining FM utility and effectiveness (for example, by using data masking techniques, Amazon Comprehend PII detection, anonymization strategies for sensitive information, Amazon Bedrock guardrails).

11.1 IAM for Bedrock ★★★ Core

Know how to secure Bedrock access with IAM. Understand service roles, identity-based policies for model access, and fine-grained permissions for knowledge bases and agents. Least-privilege API access to FMs is called out directly (Skill 2.3.3).

Core docs

Security in Amazon Bedrock — security overview and best practices
How Bedrock works with IAM — permissions for models, knowledge bases, agents
Bedrock service roles — service roles and trust policies

11.2 Data protection and encryption ★★ Important

Know how data is protected in Bedrock. Understand encryption at rest (AWS KMS), encryption in transit (TLS), VPC endpoints for private connectivity, and data-retention controls (for example, S3 Lifecycle configurations). Amazon Bedrock does not use your inputs or outputs to train the base FMs.

Core docs

Data protection in Bedrock — how data is handled and protected
Data encryption — KMS integration for custom models and resources
VPC endpoints — private connectivity without internet
Use interface VPC endpoints (AWS PrivateLink) — private API access

11.3 PII protection and network isolation ★★ Important

Know how to keep sensitive data private during FM interactions (Skills 3.2.1 to 3.2.3): detect PII with Amazon Comprehend and Amazon Macie, redact with guardrails, apply masking and anonymization, isolate networks with VPC endpoints, and govern granular data access with AWS Lake Formation.

Core docs

Amazon Macie — discover PII in S3 knowledge-base data
Amazon Comprehend PII detection — identify personal information in text
AWS Lake Formation — granular access control for data
Lambda VPC configuration — run Lambda in a VPC for private Bedrock access

Chapter 12 — AI governance and compliance

Maps to Task Statement 3.3 — Implement AI governance and compliance mechanisms

Skill 3.3.1 — Develop compliance frameworks to ensure regulatory compliance for FM deployments (for example, by using SageMaker AI to develop programmatic model cards, AWS Glue to automatically track data lineage, metadata tagging for systematic data source attribution, CloudWatch Logs to collect comprehensive decision logs).
Skill 3.3.2 — Implement data source tracking to maintain traceability in GenAI applications (for example, by using AWS Glue Data Catalog to register data sources, metadata tagging for source attribution in FM-generated content, CloudTrail for audit logging).
Skill 3.3.3 — Create organizational governance systems to ensure consistent oversight of FM implementations (for example, by using comprehensive frameworks that align with organizational policies, regulatory requirements, and responsible AI principles).
Skill 3.3.4 — Implement continuous monitoring and advanced governance controls to support safety audits and regulatory readiness (for example, by using automated detection for misuse, drift, and policy violations, bias drift monitoring, automated alerting and remediation workflows, token-level redaction, response logging, AI output policy filters).

12.1 Model invocation logging ★★ Important

Know how to enable and use invocation logging for auditing and compliance (Skills 3.3.1, 3.3.4). Understand what is logged (prompts, responses, metadata), the destinations (S3, CloudWatch Logs), and how CloudTrail records API activity.

Core docs

Model invocation logging — log prompts and responses to S3 and CloudWatch for auditing
Monitor Bedrock — logging and monitoring overview
CloudTrail for Bedrock — API activity logging for compliance

12.2 Data lineage, traceability, and compliance ★ Light

Know how to maintain traceability and meet compliance requirements (Skills 3.3.1 to 3.3.3): register data sources in the AWS Glue Data Catalog, tag metadata for source attribution, collect decision logs in CloudWatch Logs, and use CloudTrail for audit. Know Bedrock's compliance posture (SOC, ISO, HIPAA eligibility, FedRAMP) and the shared responsibility model. Bedrock does not use your data to train the base foundation models.

Core docs

Compliance validation for Bedrock — SOC, ISO, HIPAA eligibility, FedRAMP
AWS Glue Data Catalog — register data sources for lineage and attribution
AWS Shared Responsibility Model — AWS vs customer security responsibilities

Chapter 13 — Responsible AI

Maps to Task Statement 3.4 — Implement responsible AI principles

Skill 3.4.1 — Develop transparent AI systems in FM outputs (for example, by using reasoning displays to provide user-facing explanations, CloudWatch to collect confidence metrics and quantify uncertainty, evidence presentation for source attribution, Amazon Bedrock agent tracing to provide reasoning traces).
Skill 3.4.2 — Apply fairness evaluations to ensure unbiased FM outputs (for example, by using pre-defined fairness metrics in CloudWatch, Amazon Bedrock Prompt Management and Amazon Bedrock Prompt Flows to perform systematic A/B testing, Amazon Bedrock with LLM-as-a-judge solutions to perform automated model evaluations).
Skill 3.4.3 — Develop policy-compliant AI systems to ensure adherence to responsible AI practices (for example, by using Amazon Bedrock guardrails based on policy requirements, model cards to document FM limitations, Lambda functions to perform automated compliance checks).

13.1 Responsible AI principles and transparency ★★ Important

Know AWS's responsible AI dimensions and how to implement them (Skills 3.4.1, 3.4.3): transparency through reasoning traces and source attribution, AI Service Cards for documented limitations, model cards, guardrails aligned to policy, and human oversight. Agent tracing provides user-facing reasoning traces.

Core docs

Responsible AI — AWS principles and practices
AWS responsible AI resources and service cards — transparency documentation for AI services
Agent trace — step-by-step agent reasoning traces
Human review with A2I — human-in-the-loop for sensitive predictions

13.2 Fairness evaluation and LLM-as-a-judge ★★ Important

Know how to evaluate fairness and quality without exhaustive human review (Skill 3.4.2): systematic A/B testing with Prompt Management and Flows, pre-defined fairness metrics in CloudWatch, and LLM-as-a-judge — using a foundation model to score another model's outputs automatically in Bedrock model evaluations.

Core docs

LLM-as-a-judge model evaluation — automated quality and fairness scoring with a judge model
SageMaker Clarify bias detection — detect bias in data and models

Part IV — Domain 4: Operational Efficiency and Optimization for GenAI Applications (12%)

Cost optimization and resource efficiency, application performance, and monitoring for generative AI applications. Three official tasks live here.

Chapter 14 — Cost optimization and resource efficiency

Maps to Task Statement 4.1 — Implement cost optimization and resource efficiency strategies

Skill 4.1.1 — Develop token efficiency systems to reduce FM costs while maintaining effectiveness (for example, by using token estimation and tracking, context window optimization, response size controls, prompt compression, context pruning, response limiting).
Skill 4.1.2 — Create cost-effective model selection frameworks (for example, by using cost-capability tradeoff evaluation, tiered FM usage based on query complexity, inference cost balancing against response quality, price-to-performance ratio measurement, efficient inference patterns).
Skill 4.1.3 — Develop high-performance FM systems to maximize resource utilization and throughput for GenAI workloads (for example, by using batching strategies, capacity planning, utilization monitoring, auto-scaling configurations, provisioned throughput optimization).
Skill 4.1.4 — Create intelligent caching systems to reduce costs and improve response times by avoiding unnecessary FM invocations (for example, by using semantic caching, result fingerprinting, edge caching, deterministic request hashing, prompt caching).

14.1 Bedrock pricing and cost management ★★ Important

Know Bedrock pricing models: on-demand (per token), provisioned throughput (reserved capacity), and batch inference (lower cost for non-real-time). Understand how to estimate costs from input and output tokens, tiered model selection by query complexity, and cost allocation tags for tracking spend.

Core docs

Amazon Bedrock pricing — on-demand, provisioned, batch pricing
Provisioned throughput — reserved capacity for predictable workloads
Batch inference — lower cost for non-real-time processing
Cost allocation tags — track costs by project, team, environment

14.2 Token optimization and caching ★★ Important

Know techniques to reduce token usage and avoid redundant calls (Skills 4.1.1, 4.1.4): prompt optimization and compression, response length limits, context pruning, and caching — including Bedrock prompt caching for repeated prefixes and semantic caching to reuse answers to similar queries.

Core docs

Inference parameters — tune quality versus cost, including max tokens
Prompt caching — reuse context for repeated prefixes

Chapter 15 — Application performance

Maps to Task Statement 4.2 — Optimize application performance

Skill 4.2.1 — Create responsive AI systems to address latency-cost tradeoffs and improve the user experience with FMs (for example, by using pre-computation to perform predictable queries, latency-optimized Amazon Bedrock models for time-sensitive applications, parallel requests for complex workflows, response streaming, performance benchmarking).
Skill 4.2.2 — Enhance retrieval performance to improve the relevance and speed of retrieved information for FM context augmentation (for example, by using index optimization, query preprocessing, hybrid search implementation with custom scoring).
Skill 4.2.3 — Implement FM throughput optimization to address the specific throughput challenges of GenAI workloads (for example, by using token processing optimization, batch inference strategies, concurrent model invocation management).
Skill 4.2.4 — Enhance FM performance to achieve optimal results for specific GenAI use cases (for example, by using model-specific parameter configurations, A/B testing to evaluate improvements, appropriate temperature and top-k/top-p selection based on requirements).
Skill 4.2.5 — Create efficient resource allocation systems specifically for FM workloads (for example, by using capacity planning for token processing requirements, utilization monitoring for prompt and completion patterns, auto-scaling configurations that are optimized for GenAI traffic patterns).
Skill 4.2.6 — Optimize FM system performance for GenAI workflows (for example, by using API call profiling for prompt-completion patterns, vector database query optimization for retrieval augmentation, latency reduction techniques specific to LLM inference, efficient service communication patterns).

15.1 Latency optimization ★★ Important

Know techniques to reduce latency (Skill 4.2.1): streaming responses, latency-optimized Bedrock models for time-sensitive apps, provisioned throughput for consistent performance, model selection by latency requirement, parallel requests, and cross-region inference for global availability.

Core docs

InvokeModelWithResponseStream API — reduce time to first token
Provisioned throughput — consistent latency for production workloads
Cross-region inference — automatic routing for global availability
Supported foundation models — pick smaller models for lower latency

15.2 Throughput and scaling patterns ★★ Important

Know how to scale generative AI applications (Skills 4.2.3, 4.2.5): batch inference and concurrent invocation management, Lambda concurrency and provisioned concurrency, API Gateway throttling, AWS Auto Scaling, and queue-based decoupling with SQS for bursts.

Core docs

Batch inference — high-throughput, non-real-time processing
Lambda concurrency — reserved vs unreserved concurrency limits
Provisioned concurrency — eliminate cold starts
API Gateway throttling — rate limiting and burst control
SQS for decoupling — queue-based architecture for burst handling

Chapter 16 — Monitoring

Maps to Task Statement 4.3 — Implement monitoring systems for GenAI applications

Skill 4.3.1 — Create holistic observability systems to provide complete visibility into FM application performance (for example, by using operational metrics, performance tracing, FM interaction tracing, business impact metrics with custom dashboards).
Skill 4.3.2 — Implement comprehensive GenAI monitoring systems to proactively identify issues and evaluate key performance indicators specific to FM implementations (for example, by using CloudWatch to track token usage; prompt effectiveness; hallucination rates; and response quality, anomaly detection for token burst patterns and response drift, Amazon Bedrock Model Invocation Logs to perform detailed request and response analysis, performance benchmarks, cost anomaly detection).
Skill 4.3.3 — Develop integrated observability solutions to provide actionable insights for FM applications (for example, by using operational metric dashboards, business impact visualizations, compliance monitoring, forensic traceability and audit logging, user interaction tracking, model behavior pattern tracking).
Skill 4.3.4 — Create tool performance frameworks to ensure optimal tool operation and utilization for FMs (for example, by using call pattern tracking, performance metric collection, tool calling observability and multi-agent coordination tracking, usage baselines for anomaly detection).
Skill 4.3.5 — Create vector store operational management systems to ensure optimal vector store operation and reliability for FM augmentation (for example, by using performance monitoring for vector databases, automated index optimization routines, data quality validation processes).
Skill 4.3.6 — Develop FM-specific troubleshooting frameworks to identify unique GenAI failure modes that are not present in traditional ML systems (for example, by using golden datasets to detect hallucinations, output diffing techniques to conduct response consistency analysis, reasoning path tracing to identify logical errors, specialized observability pipelines).

16.1 CloudWatch metrics for Bedrock ★★ Important

Know the key CloudWatch metrics for Bedrock (invocation count, latency, throttling, errors, token usage) and GenAI-specific KPIs (prompt effectiveness, hallucination rates). Understand dashboards, cost anomaly detection, and using model invocation logs for detailed request and response analysis.

Core docs

Monitor Amazon Bedrock — metrics and logging overview
Runtime metrics — invocation count, latency, throttling, token usage
CloudWatch alarms — alerting on metric thresholds
CloudWatch dashboards — visualise GenAI application health

16.2 Distributed tracing and tool observability ★ Light

Know how to trace requests through generative AI applications (Skills 4.3.1, 4.3.4): X-Ray for distributed tracing across Lambda and API Gateway, plus tool-calling and multi-agent coordination observability, with usage baselines for anomaly detection.

Core docs

AWS X-Ray — distributed tracing for GenAI pipelines
X-Ray with Lambda — trace Lambda execution time
Agent trace — inspect agent reasoning and tool calls

Part V — Domain 5: Testing, Validation, and Troubleshooting (11%)

Evaluation systems and troubleshooting for generative AI applications. Two official tasks live here.

Chapter 17 — Evaluation systems

Maps to Task Statement 5.1 — Implement evaluation systems for GenAI

Skill 5.1.1 — Develop comprehensive assessment frameworks to evaluate the quality and effectiveness of FM outputs beyond traditional ML evaluation approaches (for example, by using metrics for relevance, factual accuracy, consistency, and fluency).
Skill 5.1.2 — Create systematic model evaluation systems to identify optimal configurations (for example, by using Amazon Bedrock Model Evaluations, A/B testing and canary testing of FMs, multi-model evaluation, cost-performance analysis to measure token efficiency, latency-to-quality ratios, and business outcomes).
Skill 5.1.3 — Develop user-centered evaluation mechanisms to continuously improve FM performance based on user experience (for example, by using feedback interfaces, rating systems for model outputs, annotation workflows to assess response quality).
Skill 5.1.4 — Create systematic quality assurance processes to maintain consistent performance standards for FMs (for example, by using continuous evaluation workflows, regression testing for model outputs, automated quality gates for deployments).
Skill 5.1.5 — Develop comprehensive assessment systems to ensure thorough evaluation from multiple perspectives for FM outputs (for example, by using RAG evaluation, automated quality assessment with LLM-as-a-Judge techniques, human feedback collection interfaces).
Skill 5.1.6 — Implement retrieval quality testing to evaluate and optimize information retrieval components for FM augmentation (for example, by using relevance scoring, context matching verification, retrieval latency measurements).
Skill 5.1.7 — Develop agent performance frameworks to ensure that agents perform tasks correctly and efficiently (for example, by using task completion rate measurements, tool usage effectiveness evaluations, Amazon Bedrock Agent evaluations, reasoning quality assessment in multi-step workflows).
Skill 5.1.8 — Create comprehensive reporting systems to communicate performance metrics and insights effectively to stakeholders for FM implementations (for example, by using visualization tools, automated reporting mechanisms, model comparison visualizations).
Skill 5.1.9 — Create deployment validation systems to maintain reliability during FM updates (for example, by using synthetic user workflows, AI-specific output validation for hallucination rates and semantic drift, automated quality checks to ensure response consistency).

17.1 Model, RAG, and agent evaluation ★★ Important

Know Bedrock evaluations end to end (Skills 5.1.2, 5.1.5, 5.1.7): automatic model evaluation, human evaluation, LLM-as-a-judge, RAG (knowledge base) evaluation for retrieval quality, and agent evaluations for task completion and reasoning quality. Know when to use automatic versus human evaluation.

Core docs

Model evaluation in Bedrock — compare models and measure quality
Automatic evaluation — programmatic quality assessment
Human evaluation — subjective quality ratings with reviewers
LLM-as-a-judge evaluation — automated scoring with a judge model
Knowledge base (RAG) evaluation — measure retrieval and generation quality

17.2 Evaluation metrics and golden datasets ★★ Important

Know the metrics for generative AI (Skills 5.1.1, 5.1.6): relevance, factual accuracy, consistency, and fluency; ROUGE and BERTScore for automatic assessment; retrieval relevance scoring and context-matching for RAG; and golden datasets used to detect hallucinations and regressions during updates.

Core docs

Evaluation metrics — metric definitions and selection
Automatic evaluation — ROUGE, BERTScore, accuracy
Knowledge base evaluation metrics — retrieval relevance and generation quality

Chapter 18 — Troubleshooting

Maps to Task Statement 5.2 — Troubleshoot GenAI applications

Skill 5.2.1 — Resolve content handling issues to ensure that necessary information is processed completely in FM interactions (for example, by using context window overflow diagnostics, dynamic chunking strategies, prompt design optimization, truncation-related error analysis).
Skill 5.2.2 — Diagnose and resolve FM integration issues to identify and fix API integration problems specific to GenAI services (for example, by using error logging, request validation, response analysis).
Skill 5.2.3 — Troubleshoot prompt engineering problems to improve FM response quality and consistency beyond basic prompt adjustments (for example, by using prompt testing frameworks, version comparison, systematic refinement).
Skill 5.2.4 — Troubleshoot retrieval system issues to identify and resolve problems that affect information retrieval effectiveness for FM augmentation (for example, by using model response relevance analysis, embedding quality diagnostics, drift monitoring, vectorization issue resolution, chunking and preprocessing remediation, vector search performance optimization).
Skill 5.2.5 — Troubleshoot prompt maintenance issues to continuously improve the performance of FM interactions (for example, by using template testing and CloudWatch Logs to diagnose prompt confusion, X-Ray to implement prompt observability pipelines, schema validation to detect format inconsistencies, systematic prompt refinement workflows).

18.1 Common issues and solutions ★★ Important

Know how to diagnose and fix common issues: throttling (increase provisioned throughput or implement backoff), context window overflow and truncation (dynamic chunking, prompt design), poor output quality (prompt engineering, model selection), retrieval problems (embedding quality, chunking, drift), and agent failures (action group and reasoning issues).

Core docs

Bedrock API common errors — error codes and meanings
Quotas for Amazon Bedrock — throttling limits and how to raise them
How retrieval works — diagnose chunking and retrieval problems

18.2 Debugging techniques ★ Light

Know debugging approaches (Skills 5.2.2, 5.2.5): model invocation logging to inspect inputs and outputs, agent trace to follow reasoning steps, X-Ray for prompt observability pipelines, schema validation for format inconsistencies, and CloudWatch Logs Insights for analysis.

Core docs

Model invocation logging — inspect prompts and responses
Agent trace — step-by-step agent reasoning
CloudWatch Logs Insights — query and analyse log data

Chapter 19 — Supporting services reference

Additional in-scope AWS services commonly used with generative AI applications. These map across the domains above; use this as a quick reference. Services are grouped by function. Only services on the exam guide's in-scope list are included.

19.1 Amazon SageMaker AI ★★ Important

Know when to use SageMaker AI versus Bedrock. Understand SageMaker JumpStart for deploying foundation models, endpoints for custom model hosting, the Model Registry for versioning customized models, and SageMaker Clarify for bias detection.

Core docs

Amazon SageMaker AI — build, train, and deploy ML models
SageMaker JumpStart — pre-trained FMs and quick deployment
SageMaker endpoints — host custom models for inference
SageMaker Clarify — bias detection and explainability

FAQ

Amazon SageMaker FAQ — pricing, supported frameworks, deployment options

19.2 Amazon Kendra ★ Light

Know Amazon Kendra as an enterprise search service that can act as a retriever for RAG. Understand when to use Kendra versus Bedrock Knowledge Bases.

Core docs

What is Amazon Kendra? — intelligent enterprise search service
Kendra data sources — supported connectors and configuration

FAQ

Amazon Kendra FAQ — pricing, document types, language support

19.3 Amazon Comprehend ★ Light

Know Amazon Comprehend for NLP tasks: sentiment analysis, entity recognition, language detection, and PII detection. Understand when to use Comprehend versus foundation models, and its role in intent recognition and pre-processing filters.

Core docs

What is Amazon Comprehend? — NLP service for text analysis
Sentiment analysis — positive, negative, neutral, mixed
Entity recognition — extract people, places, organisations
PII detection — identify personal information in text

FAQ

Amazon Comprehend FAQ — supported languages, custom models

19.4 Amazon Textract and Rekognition ★ Light

Know these services for multimodal AI applications. Textract extracts text, tables, and forms from documents. Rekognition provides image and video analysis (object detection, facial analysis, content moderation).

Core docs

Amazon Textract — extract text, tables, forms from documents
Amazon Rekognition — image and video analysis, content moderation

FAQ

Amazon Textract FAQ — supported formats, accuracy, pricing
Amazon Rekognition FAQ — face detection, custom labels, video analysis

19.5 Amazon Transcribe ★ Light

Know Amazon Transcribe for speech-to-text conversion. Understand real-time versus batch transcription, custom vocabularies, and integration with generative AI for voice-enabled applications.

Core docs

What is Amazon Transcribe? — speech-to-text conversion
Real-time transcription — live audio to text
Custom vocabularies — improve accuracy for domain terms

FAQ

Amazon Transcribe FAQ — languages, accuracy, medical transcription

19.6 Amazon Augmented AI (A2I) ★ Light

Know A2I for human-in-the-loop workflows. Understand when to route AI predictions to human review, creating review workflows, and integrating human oversight for quality assurance.

Core docs

What is Amazon A2I? — human review for ML predictions
Creating human review workflows — configure review triggers and routing

Chapter 20 — Data and analytics services

Supporting services for data processing and analytics in GenAI applications.

20.1 AWS Glue ★ Light

Know AWS Glue for ETL, data quality, and lineage. Understand the Glue Data Catalog for metadata and source attribution, Glue crawlers for schema discovery, Glue Data Quality for validation, and Glue jobs for transforming data before ingestion into knowledge bases.

Core docs

What is AWS Glue? — serverless data integration and ETL
Glue Data Catalog — metadata repository and source attribution
Glue ETL jobs — transform data for AI workloads
Glue Data Quality — validate data before FM consumption

FAQ

AWS Glue FAQ — pricing, supported sources, Spark version

20.2 Amazon Athena ★ Light

Know Athena for serverless SQL queries on S3 data. Useful for analyzing model invocation logs, querying data lakes, and preparing data for RAG applications.

Core docs

What is Amazon Athena? — serverless SQL queries on S3
Querying data — analyse logs and prepare RAG data
Athena with the Glue Data Catalog — shared metadata for tables

FAQ

Amazon Athena FAQ — pricing model, supported formats

20.3 Amazon Kinesis and Amazon Data Firehose ★ Light

Know Kinesis for real-time data streaming. Understand Kinesis Data Streams for real-time ingestion and Amazon Data Firehose (formerly Kinesis Data Firehose) for delivery to destinations, plus use cases like real-time inference pipelines.

Core docs

Amazon Kinesis Data Streams — real-time data ingestion
Amazon Data Firehose — load streaming data to destinations
Kinesis with Lambda — process streams with Lambda

FAQ

Amazon Kinesis FAQ — throughput, retention, scaling

20.4 Amazon EMR ★ Light

Know EMR for big data processing. Understand EMR for large-scale data preparation, distributed processing of training or ingestion data, and integration with SageMaker.

Core docs

What is Amazon EMR? — managed big data processing
EMR with Spark — large-scale data transformation

FAQ

Amazon EMR FAQ — instance types, pricing, supported frameworks

20.5 Amazon QuickSight (Amazon Quick Suite) ★ Light

Know Amazon QuickSight, now part of Amazon Quick Suite, for business intelligence. Understand natural-language querying and how to visualize GenAI application metrics.

Core docs

Amazon Quick Suite (QuickSight) — serverless BI, visualisation, and natural-language querying

FAQ

Amazon QuickSight FAQ — pricing, data sources, embedding

20.6 Amazon MSK ★ Light

Know Amazon MSK (Managed Streaming for Apache Kafka) for event streaming. Understand when to use MSK versus Kinesis for GenAI data pipelines.

Core docs

What is Amazon MSK? — managed Apache Kafka

FAQ

Amazon MSK FAQ — Kafka versions, scaling, pricing

Chapter 21 — Application integration services

Services for building event-driven and workflow-based GenAI applications.

21.1 AWS Step Functions ★★ Important

Know Step Functions for orchestrating GenAI workflows: ReAct and reasoning patterns, review and approval processes, circuit breakers and stopping conditions, native Bedrock integration, error handling with Retry and Catch, and Standard versus Express workflows.

Core docs

What is AWS Step Functions? — serverless workflow orchestration
Step Functions with Bedrock — native Bedrock integration
Standard vs Express workflows — long-running vs high-volume
Error handling — Retry, Catch, fallback patterns

FAQ

AWS Step Functions FAQ — pricing, state limits, integrations

21.2 Amazon EventBridge ★ Light

Know EventBridge for event-driven architectures. Understand event buses, rules, targets, and how to trigger GenAI workflows based on events.

Core docs

What is Amazon EventBridge? — serverless event bus
Event patterns — filter and route events
Rules and targets — trigger GenAI workflows from events

FAQ

Amazon EventBridge FAQ — event size, delivery guarantees

Know SNS for pub/sub messaging and SQS for queue-based decoupling. Understand fan-out patterns, dead-letter queues, and how to build resilient GenAI pipelines.

Core docs

Amazon SNS — pub/sub messaging and notifications
Amazon SQS — message queuing for decoupling
SNS to SQS fanout — broadcast to multiple queues
SQS dead-letter queues — handle failed messages

FAQ

Amazon SNS FAQ — delivery, filtering, pricing
Amazon SQS FAQ — visibility timeout, message retention

21.4 Amazon AppFlow ★ Light

Know AppFlow for SaaS data integration. Understand how to bring data from Salesforce, ServiceNow, Slack, and others into S3 for use with Bedrock Knowledge Bases.

Core docs

What is Amazon AppFlow? — SaaS data integration
Supported sources — Salesforce, ServiceNow, Slack connectors
Creating flows — configure data transfers to S3

FAQ

Amazon AppFlow FAQ — connectors, scheduling, data transformation

21.5 AWS AppConfig ★ Light

Know AppConfig for feature flags and dynamic configuration. Understand deployment strategies and how to use AppConfig for dynamic model selection and A/B testing prompt variations (Skill 1.2.2).

Core docs

What is AWS AppConfig? — dynamic configuration management
Feature flags — toggle features without deploys
Deployment strategies — gradual rollouts, A/B testing prompts

Chapter 22 — Compute and container services

Compute options for hosting GenAI applications.

22.1 AWS Lambda ★★ Important

Lambda is the primary compute for serverless GenAI. Know function configuration, environment variables, VPC access, response streaming, and integration with Bedrock. Lambda also hosts stateless MCP servers for lightweight tool access (Skill 2.1.7).

Core docs

AWS Lambda — serverless compute for GenAI backends
Invoke a model with the Bedrock API — call models from Lambda
Lambda environment variables — store configuration securely
Lambda VPC access — private network connectivity
Response streaming — stream large responses

FAQ

AWS Lambda FAQ — timeout limits, memory, pricing

22.2 Amazon EC2 ★ Light

Know EC2 for GenAI workloads requiring GPUs or sustained compute. Understand accelerated-computing instance families, Spot instances for cost optimization, and when to use EC2 versus serverless.

Core docs

Amazon EC2 — virtual servers for sustained workloads
Accelerated computing instances — GPU instances for model inference
Spot instances — cost savings for interruptible workloads

FAQ

Amazon EC2 FAQ — instance types, pricing, networking

22.3 Container services (ECS, EKS, Fargate) ★ Light

Know container options for GenAI applications. Understand ECS versus EKS, Fargate for serverless containers, ECR for image storage, and hosting heavier MCP servers on ECS (Skill 2.1.7).

Core docs

Amazon ECS — managed container orchestration
Amazon EKS — managed Kubernetes
AWS Fargate — serverless containers
Amazon ECR — container image registry

FAQ

Amazon ECS FAQ — task definitions, networking modes
Amazon EKS FAQ — Kubernetes versions, add-ons

22.4 AWS App Runner ★ Light

Know App Runner for fully managed container deployments. Understand when to use App Runner for simple GenAI APIs without managing infrastructure.

Core docs

What is AWS App Runner? — fully managed container deployment
Creating services — deploy from source or image

FAQ

AWS App Runner FAQ — auto-scaling, custom domains

Chapter 23 — Database services

Database options for GenAI applications.

23.1 Amazon DynamoDB ★★ Important

Know DynamoDB for GenAI application state: conversation history storage (Skill 1.6.2), session state, and DynamoDB Streams for change capture to trigger GenAI workflows.

Core docs

Amazon DynamoDB — serverless NoSQL for session and conversation state
DynamoDB Streams — trigger GenAI on data changes
Best practices — key design, indexing patterns

FAQ

Amazon DynamoDB FAQ — pricing, consistency, transactions

23.2 Amazon Aurora and RDS ★★ Important

Know Aurora PostgreSQL with the pgvector extension as a managed vector store for Bedrock Knowledge Bases, and Amazon RDS paired with S3 document repositories (Skill 1.4.1).

Core docs

Amazon Aurora — managed relational database
Aurora with pgvector — PostgreSQL extension for vector search

FAQ

Amazon Aurora FAQ — pricing, scaling, compatibility

23.3 Amazon ElastiCache ★ Light

Know ElastiCache for caching GenAI responses. Understand Redis-compatible semantic caching, session management, and reducing Bedrock API costs through response caching (Skill 4.1.4).

Core docs

Amazon ElastiCache — in-memory caching for responses
Caching strategies — reduce Bedrock API costs

FAQ

Amazon ElastiCache FAQ — node types, replication, pricing

23.4 Amazon DocumentDB and Neptune ★ Light

Know DocumentDB for document storage with vector search, and Amazon Neptune (including Neptune Analytics) as a graph and vector option for RAG. Understand these as alternatives to OpenSearch for specific data shapes.

Core docs

Amazon DocumentDB — MongoDB-compatible document database
DocumentDB vector search — alternative to OpenSearch for RAG
Amazon Neptune Analytics — graph analytics with vector similarity

FAQ

Amazon DocumentDB FAQ — MongoDB compatibility, scaling

Chapter 24 — Developer tools and CI/CD

Tools for building, deploying, and managing GenAI applications.

24.1 AWS CloudFormation and CDK ★ Light

Know CloudFormation for infrastructure as code and CDK for defining infrastructure in programming languages. Understand deploying Bedrock resources (agents, knowledge bases, guardrails) with these tools.

Core docs

AWS CloudFormation — infrastructure as code
AWS CDK — define infrastructure in code
Bedrock CloudFormation resources — agents, knowledge bases, guardrails

FAQ

AWS CloudFormation FAQ — stack limits, drift detection
AWS CDK FAQ — supported languages, versioning

24.2 CI/CD services (CodePipeline, CodeBuild, CodeDeploy) ★ Light

Know AWS CI/CD services for deploying GenAI applications with automated testing, security scans, and rollback (Skill 2.3.5). Understand CodePipeline for orchestration, CodeBuild for building and testing, and CodeDeploy for deployment strategies.

Core docs

AWS CodePipeline — continuous delivery orchestration
AWS CodeBuild — build and test automation
AWS CodeDeploy — deployment automation
AWS CodeArtifact — artifact repository

FAQ

AWS CodePipeline FAQ — triggers, actions, integrations
AWS CodeBuild FAQ — build environments, caching

24.3 AWS Amplify ★ Light

Know Amplify for full-stack GenAI applications. Understand the Amplify AI kit for integrating Bedrock into web and mobile applications (Skill 2.5.2).

Core docs

AWS Amplify — full-stack web and mobile apps
Amplify AI kit — integrate Bedrock into Amplify apps

FAQ

AWS Amplify FAQ — hosting, authentication, data

Chapter 25 — Networking and content delivery

Networking for GenAI applications.

25.1 Amazon API Gateway ★★ Important

Know API Gateway for fronting FMs: request validation, throttling and rate limiting, streaming and chunked transfer encoding, WebSocket APIs for real-time interaction, and request transformations for model routing (Skills 2.4.x, 2.5.1).

Core docs

Amazon API Gateway — managed API front door
Request throttling — rate limiting and burst control
WebSocket APIs — bidirectional real-time communication

FAQ

Amazon API Gateway FAQ — API types, throttling, pricing

25.2 CloudFront, ELB, and Global Accelerator ★ Light

Know CloudFront for caching and global distribution (including edge caching for responses), ELB for distributing traffic to GenAI backends, and Global Accelerator for improved global performance.

Core docs

Amazon CloudFront — global content delivery network
Elastic Load Balancing — distribute traffic across backends
AWS Global Accelerator — improved global performance

FAQ

Amazon CloudFront FAQ — caching, pricing, origins

25.3 Amazon Route 53 ★ Light

Know Route 53 for DNS management. Understand routing policies and health checks for high-availability GenAI applications.

Core docs

Amazon Route 53 — DNS management and traffic routing
Routing policies — weighted, latency-based, geolocation

FAQ

Amazon Route 53 FAQ — health checks, domain registration

Chapter 26 — Security services

Additional security services for GenAI applications.

26.1 Amazon Cognito ★★ Important

Know Cognito for user authentication in GenAI applications and for identity federation between FM services and enterprise systems (Skill 2.3.3). Understand user pools, identity pools, and token-based authentication for API Gateway and Bedrock access.

Core docs

Amazon Cognito — user authentication for GenAI apps
User pools — user directory and sign-up/sign-in
Identity pools — temporary AWS credentials for users

FAQ

Amazon Cognito FAQ — pricing, MFA, federation

26.2 AWS Secrets Manager ★ Light

Know Secrets Manager for storing API keys and credentials used in GenAI applications. Understand automatic rotation and integration with Lambda.

Core docs

AWS Secrets Manager — store API keys and credentials securely
Automatic rotation — rotate secrets without downtime

FAQ

AWS Secrets Manager FAQ — pricing, Lambda integration

26.3 AWS WAF ★ Light

Know WAF for protecting GenAI APIs from attacks. Understand rate limiting, managed rules, and integration with API Gateway.

Core docs

AWS WAF — web application firewall
Rate-based rules — protect against abuse
WAF with API Gateway — protect GenAI APIs

FAQ

AWS WAF FAQ — managed rules, pricing, logging

26.4 Amazon Macie ★ Light

Know Macie for discovering and protecting sensitive data. Understand how to use Macie to identify PII in S3 buckets used for RAG knowledge bases (Skill 3.2.2).

Core docs

Amazon Macie — discover PII in S3 knowledge bases
Discovering sensitive data — automated sensitive data detection

FAQ

Amazon Macie product page — supported data types, pricing

26.5 AWS Encryption SDK ★ Light

Know the Encryption SDK for client-side encryption. Understand envelope encryption and when to use client-side encryption for sensitive GenAI data.

Core docs

AWS Encryption SDK — client-side encryption library
How it works — envelope encryption with KMS

Chapter 27 — Storage services

Storage options for GenAI data.

27.1 Amazon S3 ★★ Important

S3 is the primary storage for GenAI data. Know bucket policies, encryption, lifecycle policies (for data-retention controls, Skill 3.2.2), and S3 as a data source for Bedrock Knowledge Bases.

Core docs

Amazon S3 — primary storage for GenAI data
Bucket policies — access control for knowledge base data
Encryption — SSE-S3, SSE-KMS, SSE-C options
Lifecycle policies — manage data retention and archival
S3 Intelligent-Tiering — automatic cost optimisation

FAQ

Amazon S3 FAQ — pricing, storage classes, durability

27.2 Amazon EBS and EFS ★ Light

Know EBS for EC2 block storage and EFS for shared file storage. Understand when to use these for GenAI workloads (model weights, working data).

Core docs

Amazon EBS — block storage for EC2 model hosting
Amazon EFS — shared file storage for model weights

FAQ

Amazon EBS FAQ — volume types, IOPS, snapshots
Amazon EFS product page — throughput modes, pricing

Chapter 28 — Management and governance

Operational services for GenAI applications.

28.1 AWS Systems Manager ★ Light

Know Systems Manager for operational management. Understand Parameter Store for configuration and Session Manager for secure access.

Core docs

AWS Systems Manager — operational management
Parameter Store — configuration and secrets (simpler than Secrets Manager)

FAQ

AWS Systems Manager FAQ — capabilities, pricing

28.2 Cost and usage analysis ★ Light

Know AWS Cost Explorer and AWS Cost Anomaly Detection for analyzing GenAI spend. Understand cost allocation tags and anomaly detection for managing Bedrock costs (Skill 4.3.2). (Note: AWS Budgets is out of scope for this exam.)

Core docs

AWS Cost Explorer — analyse and visualise Bedrock costs
AWS Cost Anomaly Detection — detect unexpected GenAI spend
Cost allocation tags — track costs by project or team

28.3 Amazon Managed Grafana ★ Light

Know Managed Grafana for advanced visualization of GenAI metrics. Understand dashboards and integration with CloudWatch.

Core docs

Amazon Managed Grafana — advanced visualisation for GenAI metrics
Data sources — connect CloudWatch, Prometheus, and more

FAQ

Amazon Managed Grafana FAQ — pricing, authentication, plugins

28.4 Amazon Q Developer in chat applications ★ Light

Know Amazon Q Developer in chat applications (formerly AWS Chatbot) for integrating AWS notifications with Slack and Microsoft Teams. Useful for GenAI application alerting and operational awareness.

Core docs

Amazon Q Developer in chat applications — AWS notifications in Slack and Teams
Getting started — configure channels and permissions

FAQ

Amazon Q Developer in chat applications FAQ — supported services, pricing