Content domains
The 5 domains of AIP-C01, with every task statement and its objectives from the official guide. Study a whole domain, or drill a single task.
1Foundation Model Integration, Data Management, and Compliance
31% of examStudy domain
1.1Analyze requirements and design GenAI solutions.13 q>
Knowledge of
- How business requirements and technical constraints translate into a GenAI architecture
- Common GenAI integration patterns and when each one fits
- The AWS Well-Architected Framework and its Generative AI Lens
- What a proof of concept should prove before a team commits to full build-out
- Reusable component design so repeated deployments stay consistent
Skills in
- Producing architectural designs that match business needs, picking suitable foundation models, integration patterns, and deployment strategies
- Building technical proofs of concept on Amazon Bedrock to test feasibility, performance, and business value ahead of a full deployment
- Standardizing technical components against the AWS Well-Architected Framework and the WA Tool Generative AI Lens so implementations stay consistent across scenarios
1.2Select and configure FMs.13 q>
Knowledge of
- Benchmarks and capability or limitation analysis used to compare foundation models
- Abstraction layers that let an application swap models or providers without a code change
- Resilience patterns: circuit breakers, cross-Region inference, graceful degradation
- Model customization options including fine-tuning and parameter-efficient methods such as LoRA and adapters
- Model versioning, registry, promotion, rollback, and retirement workflows
Skills in
- Evaluating and choosing foundation models against business use cases using performance benchmarks and capability analysis
- Designing flexible architectures with AWS Lambda, Amazon API Gateway, and AWS AppConfig that allow dynamic model selection and provider switching without code changes
- Building resilient AI systems with AWS Step Functions circuit breaker patterns, Amazon Bedrock Cross-Region Inference, cross-Region model deployment, and graceful degradation
- Managing customized model deployment and lifecycle with Amazon SageMaker AI, LoRA and adapter techniques, SageMaker Model Registry versioning, automated pipelines, rollback strategies, and model retirement
1.3Implement data validation and processing pipelines for FM consumption.23 q>
Knowledge of
- Data quality checks and where they belong in a pipeline feeding a foundation model
- Multimodal processing needs across text, image, audio, and tabular data
- Model-specific input formats, including Amazon Bedrock request payloads and conversation structures
- Enrichment and normalization techniques that raise response quality
Skills in
- Building data validation workflows with AWS Glue Data Quality, SageMaker Data Wrangler, custom Lambda functions, and Amazon CloudWatch metrics
- Creating processing workflows for complex and multimodal data using Amazon Bedrock multimodal models, SageMaker Processing, and Amazon Transcribe
- Formatting inference input to model requirements: JSON for Amazon Bedrock API requests, structured payloads for SageMaker AI endpoints, conversation formatting for dialog apps
- Improving input quality using Amazon Bedrock to reformat text, Amazon Comprehend to extract entities, and Lambda functions to normalize data
1.4Design and implement vector store solutions.23 q>
Knowledge of
- Vector database options on AWS and their tradeoffs for FM augmentation
- Metadata frameworks that sharpen search precision and context awareness
- Indexing, sharding, and multi-index strategies for semantic search at scale
- Connectors to enterprise document stores, knowledge bases, and wikis
- Freshness mechanisms: incremental updates, change detection, scheduled refresh
Skills in
- Architecting vector databases for FM augmentation with Amazon Bedrock Knowledge Bases, Amazon OpenSearch Service with the Neural plugin, Amazon RDS alongside Amazon S3 document repositories, and Amazon DynamoDB for metadata and embeddings
- Designing metadata frameworks using S3 object metadata, custom attributes, and tagging systems for domain classification
- Tuning vector database performance with OpenSearch sharding strategies, multi-index approaches, and hierarchical indexing
- Building integration components that connect GenAI applications to document management systems, knowledge bases, and internal wikis
- Keeping vector stores current with incremental updates, real-time change detection, automated synchronization, and scheduled refresh pipelines
1.5Design retrieval mechanisms for FM augmentation.13 q>
Knowledge of
- Chunking strategies: fixed-size, hierarchical, and content-structure aware
- Embedding model selection criteria such as dimensionality, domain fit, and cost
- Vector search deployment options including OpenSearch Service, Aurora pgvector, and Bedrock Knowledge Bases
- Hybrid search and reranking to raise retrieval relevance
- Query expansion, decomposition, and transformation techniques
- Standardized retrieval access patterns such as function calling and Model Context Protocol
Skills in
- Segmenting documents using Amazon Bedrock chunking, Lambda-based fixed-size chunking, and custom hierarchical chunking driven by content structure
- Selecting and configuring embedding models such as Amazon Titan embeddings by dimensionality and domain fit, and batch-generating embeddings with Lambda
- Deploying vector search on OpenSearch Service, Amazon Aurora with pgvector, or Amazon Bedrock Knowledge Bases managed vector stores
- Improving relevance with semantic search, hybrid keyword-plus-vector search, and Amazon Bedrock reranker models
- Handling queries with Amazon Bedrock query expansion, Lambda query decomposition, and Step Functions query transformation
- Exposing retrieval through consistent interfaces: function calling, MCP clients for vector queries, and standardized retrieval APIs
1.6Implement prompt engineering strategies and governance for FM interactions.13 q>
Knowledge of
- Instruction frameworks: role definitions, response templates, guardrail-enforced guidelines
- Conversation state and intent recognition for interactive systems
- Prompt versioning, approval workflows, and audit trails
- Prompt regression testing and edge case validation
- Iterative refinement patterns including chain-of-thought and output format specs
- Prompt chaining and conditional branching for complex tasks
Skills in
- Controlling model behavior with Amazon Bedrock Prompt Management role definitions, Amazon Bedrock Guardrails, and response format templates
- Maintaining conversational context using Step Functions clarification workflows, Amazon Comprehend intent recognition, and DynamoDB conversation history
- Governing prompts with parameterized templates and approval workflows in Bedrock Prompt Management, S3 template repositories, AWS CloudTrail usage tracking, and CloudWatch Logs
- Assuring prompt quality with Lambda output verification, Step Functions edge case testing, and CloudWatch prompt regression checks
- Refining prompts iteratively using structured input components, output format specifications, chain-of-thought patterns, and feedback loops
- Designing complex prompt systems with Amazon Bedrock Prompt Flows for sequential chains, conditional branching, reusable components, and pre- and post-processing steps
2Implementation and Integration
26% of examStudy domain
2.1Implement agentic AI solutions and tool integrations.13 q>
Knowledge of
- Agent frameworks on AWS: Strands Agents, AWS Agent Squad, and Model Context Protocol
- Agent memory and state management
- Structured reasoning patterns such as ReAct and chain-of-thought
- Safety rails for autonomous loops: stopping conditions, timeouts, resource boundaries, circuit breakers
- Multi-model coordination and ensemble aggregation
- Human-in-the-loop review and feedback patterns
- Tool definition, parameter validation, and MCP server hosting options
Skills in
- Building autonomous agents with memory and state management using Strands Agents, AWS Agent Squad for multi-agent systems, and MCP for agent-tool interactions
- Implementing structured reasoning with Step Functions for ReAct patterns and chain-of-thought approaches
- Constraining agent behavior with Step Functions stopping conditions, Lambda timeouts, IAM resource boundaries, and circuit breakers
- Coordinating multiple specialized models with custom aggregation logic and model selection frameworks
- Adding human expertise through Step Functions review and approval orchestration and API Gateway feedback collection
- Integrating tools reliably with the Strands API, standardized function definitions, and Lambda-based error handling and parameter validation
- Extending model capabilities with stateless MCP servers on Lambda, complex MCP servers on Amazon ECS, and MCP client libraries for consistent access
2.2Implement model deployment strategies.23 q>
Knowledge of
- On-demand versus provisioned throughput versus self-hosted endpoint tradeoffs
- How LLM deployment differs from traditional ML: memory footprint, GPU utilization, token throughput
- Container-based serving and model loading strategies
- Model cascading and right-sizing smaller models for narrow tasks
Skills in
- Deploying foundation models to match application needs using Lambda for on-demand invocation, Amazon Bedrock provisioned throughput, and SageMaker AI endpoints for hybrid solutions
- Addressing LLM-specific deployment challenges with container patterns tuned for memory, GPU utilization, and token processing capacity, plus specialized model loading strategies
- Balancing performance against resource cost by right-sizing model choice, using smaller pre-trained models for specific tasks, and applying API-based model cascading for routine queries
2.3Design and implement enterprise integration architectures.23 q>
Knowledge of
- Legacy and event-driven integration patterns for FM capabilities
- Identity federation and least-privilege access between FM services and enterprise systems
- Hybrid and edge deployment options such as AWS Outposts and AWS Wavelength
- Data residency and jurisdictional compliance constraints
- CI/CD and GenAI gateway architectures for centralized, governed model consumption
Skills in
- Connecting FM capabilities to existing enterprise environments through API integrations with legacy systems, event-driven loose coupling, and data synchronization patterns
- Enhancing existing applications with API Gateway microservice integrations, Lambda webhook handlers, and Amazon EventBridge event-driven integrations
- Securing access with identity federation, role-based access control over models and data, and least-privilege FM API access
- Enabling cross-environment compliance using AWS Outposts for on-premises data, AWS Wavelength for edge deployments, and secure cloud-to-on-premises routing
- Building CI/CD pipelines and GenAI gateways with AWS CodePipeline, AWS CodeBuild, automated testing with security scans and rollback, centralized abstraction layers, and observability controls
2.4Implement FM API integrations.23 q>
Knowledge of
- Synchronous versus asynchronous invocation patterns for Amazon Bedrock
- Streaming response delivery over WebSockets, server-sent events, and chunked transfer encoding
- Resilience mechanisms: exponential backoff, rate limiting, fallbacks, distributed tracing
- Static versus dynamic content-based model routing
Skills in
- Building flexible model interaction using Amazon Bedrock APIs for synchronous requests, AWS SDKs with Amazon SQS for asynchronous processing, and API Gateway with request validation
- Delivering real-time output with Amazon Bedrock streaming APIs, WebSockets or server-sent events, and API Gateway chunked transfer encoding
- Making FM calls resilient with AWS SDK exponential backoff, API Gateway rate limiting, graceful degradation fallbacks, and AWS X-Ray observability across service boundaries
- Routing intelligently with static configurations, Step Functions dynamic content-based routing, metric-driven model routing, and API Gateway request transformations
2.5Implement application integration patterns and development tools.13 q>
Knowledge of
- GenAI-specific API concerns: streaming, token limits, model timeouts
- Low-code and declarative build options including AWS Amplify and Bedrock Prompt Flows
- Business system automation patterns for CRM and document processing
- Developer productivity tooling such as Amazon Q Developer
- Agent orchestration patterns and prompt chaining
- GenAI troubleshooting tooling across logs and traces
Skills in
- Designing FM API interfaces that handle streaming responses through API Gateway, token limit management, and retry strategies for model timeouts
- Accelerating adoption with AWS Amplify declarative UI components, OpenAPI-first development, and Amazon Bedrock Prompt Flows no-code workflow builders
- Enhancing business systems using Lambda for CRM enhancements, Step Functions for document processing orchestration, and Amazon Bedrock Data Automation for automated data workflows
- Improving developer productivity with Amazon Q Developer for code generation and refactoring, API assistance, AI component testing, and performance optimization
- Building advanced applications with Strands Agents and AWS Agent Squad for native orchestration, Step Functions for agent design patterns, and Amazon Bedrock prompt chaining
- Troubleshooting faster with CloudWatch Logs Insights prompt and response analysis, X-Ray tracing of FM API calls, and Amazon Q Developer error pattern recognition
3AI Safety, Security, and Governance
20% of examStudy domain
3.1Implement input and output safety controls.3 q>
Knowledge of
- Amazon Bedrock Guardrails for input filtering and response filtering
- Hallucination reduction through grounding, confidence scoring, and structured output enforcement
- Defense-in-depth layering across pre-processing, model, post-processing, and API response
- Adversarial threats: prompt injection, jailbreaks, and how they are detected
- Safety classifiers and automated adversarial testing
Skills in
- Protecting against harmful inputs with Amazon Bedrock Guardrails content filters, Step Functions and Lambda custom moderation workflows, and real-time validation
- Preventing harmful outputs with Bedrock Guardrails response filtering, specialized FM evaluations for moderation and toxicity, and deterministic text-to-SQL transformations
- Reducing hallucinations by grounding responses in Amazon Bedrock Knowledge Bases, applying confidence scoring and semantic similarity verification, and enforcing JSON Schema structured outputs
- Layering defenses with Amazon Comprehend pre-processing filters, Bedrock model-based guardrails, Lambda post-processing validation, and API Gateway response filtering
- Detecting advanced threats with prompt injection and jailbreak detection, input sanitization and content filters, safety classifiers, and automated adversarial testing workflows
3.2Implement data security and privacy controls.23 q>
Knowledge of
- Network isolation for FM workloads using VPC endpoints
- IAM and AWS Lake Formation for granular data access control
- PII detection with Amazon Comprehend and Amazon Macie
- Data retention, masking, and anonymization strategies
- Amazon Bedrock native data privacy behavior
Skills in
- Securing FM deployments with VPC endpoints for network isolation, IAM policies for data access, AWS Lake Formation for granular permissions, and CloudWatch data access monitoring
- Protecting sensitive information with Amazon Comprehend and Amazon Macie PII detection, Amazon Bedrock native privacy features, Bedrock Guardrails output filtering, and S3 Lifecycle retention policies
- Preserving user privacy without losing model utility using data masking, Comprehend PII detection, anonymization strategies, and Bedrock Guardrails
3.3Implement AI governance and compliance mechanisms.3 q>
Knowledge of
- Model cards and programmatic model documentation in SageMaker AI
- Data lineage tracking and metadata tagging for source attribution
- Audit logging with CloudTrail and CloudWatch Logs
- Organizational governance frameworks aligned to policy, regulation, and responsible AI principles
- Continuous monitoring for misuse, drift, bias drift, and policy violations
Skills in
- Meeting regulatory requirements with SageMaker AI programmatic model cards, AWS Glue automated data lineage tracking, metadata tagging for source attribution, and CloudWatch Logs decision logs
- Maintaining traceability by registering sources in the AWS Glue Data Catalog, tagging metadata for attribution in FM-generated content, and using CloudTrail audit logging
- Establishing organizational governance frameworks aligned with company policies, regulatory requirements, and responsible AI principles
- Running continuous governance controls: automated misuse, drift, and policy violation detection, bias drift monitoring, automated alerting and remediation, token-level redaction, response logging, and output policy filters
3.4Implement responsible AI principles.3 q>
Knowledge of
- Transparency mechanisms: reasoning displays, source attribution, confidence and uncertainty metrics
- Amazon Bedrock agent tracing for reasoning traces
- Fairness metrics and bias evaluation approaches
- LLM-as-a-judge automated model evaluation
- Model cards for documenting limitations and intended use
Skills in
- Making FM outputs transparent with user-facing reasoning displays, CloudWatch confidence metrics and uncertainty quantification, evidence-based source attribution, and Amazon Bedrock agent tracing
- Evaluating fairness with predefined fairness metrics in CloudWatch, systematic A/B testing via Bedrock Prompt Management and Prompt Flows, and automated evaluations using Amazon Bedrock LLM-as-a-judge
- Enforcing responsible AI policy with Bedrock Guardrails configured to policy requirements, model cards documenting FM limitations, and Lambda automated compliance checks
4Operational Efficiency and Optimization for GenAI Applications
12% of examStudy domain
4.1Implement cost optimization and resource efficiency strategies.3 q>
Knowledge of
- Token economics: estimation, tracking, context window budgeting, prompt compression
- Tiered model selection and price-to-performance evaluation
- Batching, capacity planning, auto scaling, and provisioned throughput
- Caching approaches: semantic caching, prompt caching, deterministic request hashing, edge caching
Skills in
- Reducing token spend with token estimation and tracking, context window optimization, response size controls, prompt compression, and context pruning
- Selecting cost-effective models through cost-capability tradeoff evaluation, tiered usage by query complexity, inference cost versus quality balancing, and price-to-performance measurement
- Maximizing throughput and utilization with batching strategies, capacity planning, utilization monitoring, auto-scaling configurations, and provisioned throughput optimization
- Avoiding unnecessary invocations with semantic caching, result fingerprinting, edge caching, deterministic request hashing, and prompt caching
4.2Optimize application performance.3 q>
Knowledge of
- Latency-cost tradeoffs and latency-optimized Amazon Bedrock models
- Streaming, pre-computation, and parallel request patterns
- Retrieval performance tuning: index optimization, query preprocessing, hybrid search scoring
- Inference parameter tuning including temperature and top-k/top-p
- Capacity planning and auto scaling shaped to GenAI traffic patterns
- Profiling FM API calls and vector database queries
Skills in
- Improving responsiveness with pre-computation for predictable queries, latency-optimized Bedrock models, parallel requests for complex workflows, response streaming, and performance benchmarking
- Speeding retrieval with index optimization, query preprocessing, and hybrid search with custom scoring
- Raising throughput with token processing optimization, batch inference, and concurrent model invocation management
- Tuning model behavior with model-specific parameter configurations, A/B testing, and requirement-driven temperature and top-k/top-p selection
- Allocating resources efficiently through token-aware capacity planning, prompt and completion utilization monitoring, and GenAI-shaped auto scaling
- Profiling and optimizing with prompt-completion API call profiling, vector database query optimization, LLM inference latency reduction, and efficient service communication patterns
4.3Implement monitoring systems for GenAI applications.3 q>
Knowledge of
- GenAI-specific KPIs: token usage, prompt effectiveness, hallucination rate, response quality
- Amazon Bedrock Model Invocation Logs for request and response analysis
- Anomaly detection for token bursts, response drift, and cost
- Tool calling and multi-agent coordination observability
- Vector store operational monitoring and index health
- GenAI failure modes that traditional ML monitoring misses
Skills in
- Building holistic observability across operational metrics, performance tracing, FM interaction tracing, and business impact metrics on custom dashboards
- Monitoring GenAI KPIs with CloudWatch tracking of token usage, prompt effectiveness, hallucination rates, and response quality, plus anomaly detection, Bedrock Model Invocation Logs, benchmarks, and cost anomaly detection
- Turning telemetry into insight with operational dashboards, business impact visualizations, compliance monitoring, forensic traceability and audit logging, and user and model behavior tracking
- Tracking tool performance with call pattern tracking, metric collection, tool calling and multi-agent coordination observability, and usage baselines for anomaly detection
- Operating vector stores with performance monitoring, automated index optimization routines, and data quality validation
- Diagnosing GenAI-specific failures with golden datasets for hallucination detection, output diffing for consistency analysis, reasoning path tracing, and specialized observability pipelines
5Testing, Validation, and Troubleshooting
11% of examStudy domain
5.1Implement evaluation systems for GenAI.3 q>
Knowledge of
- FM output quality metrics: relevance, factual accuracy, consistency, fluency
- Amazon Bedrock Model Evaluations, A/B testing, and canary testing
- Human feedback, rating, and annotation workflows
- Continuous evaluation, regression testing, and automated quality gates
- RAG evaluation and LLM-as-a-judge techniques
- Retrieval quality measures: relevance scoring, context matching, retrieval latency
- Agent evaluation: task completion, tool usage effectiveness, reasoning quality
- Deployment validation for hallucination rate and semantic drift
Skills in
- Assessing FM output quality with metrics for relevance, factual accuracy, consistency, and fluency that go beyond traditional ML evaluation
- Finding optimal configurations using Amazon Bedrock Model Evaluations, A/B and canary testing, multi-model evaluation, and cost-performance analysis of token efficiency, latency-to-quality ratios, and business outcomes
- Capturing user signal through feedback interfaces, output rating systems, and annotation workflows
- Maintaining standards with continuous evaluation workflows, output regression testing, and automated quality gates in deployments
- Evaluating from multiple angles with RAG evaluation, automated LLM-as-a-Judge quality assessment, and human feedback interfaces
- Testing retrieval quality with relevance scoring, context matching verification, and retrieval latency measurement
- Evaluating agents with task completion rates, tool usage effectiveness, Amazon Bedrock Agent evaluations, and reasoning quality assessment in multi-step workflows
- Reporting to stakeholders with visualization tools, automated reporting, and model comparison visualizations
- Validating deployments with synthetic user workflows, AI-specific output validation for hallucination rates and semantic drift, and automated response consistency checks
5.2Troubleshoot GenAI applications.3 q>
Knowledge of
- Context window overflow and truncation symptoms
- FM API integration failure modes and how to read them
- Prompt testing frameworks and version comparison
- Retrieval failure diagnosis: embedding quality, drift, chunking, vector search performance
- Prompt observability with CloudWatch Logs, X-Ray, and schema validation
Skills in
- Resolving content handling issues with context window overflow diagnostics, dynamic chunking, prompt design optimization, and truncation error analysis
- Diagnosing FM integration problems through error logging, request validation, and response analysis
- Fixing prompt engineering problems with prompt testing frameworks, version comparison, and systematic refinement
- Repairing retrieval systems by analyzing response relevance, diagnosing embedding quality, monitoring drift, resolving vectorization issues, remediating chunking and preprocessing, and optimizing vector search performance
- Sustaining prompt health with template testing and CloudWatch Logs to diagnose prompt confusion, X-Ray prompt observability pipelines, schema validation for format inconsistencies, and systematic refinement workflows