Advanced certification validating skills in designing, developing, and deploying generative AI applications on AWS. Covers Bedrock, RAG, and prompt engineering.
FM Selection, Comparison & Inference Parameters, Amazon Bedrock Model Catalog, Data Preprocessing & ETL Pipelines, Vector DB, Embeddings & Chunking Strategies, RAG & Bedrock Knowledge Bases, Prompt Engineering, Templates & Caching
Bedrock Agents & Action Groups, Tool Use & Multi-Agent Orchestration, Bedrock API, Converse API & Streaming, Amazon Q Developer, Model Deployment & Serverless Inference, Microservices & Event-Driven Integration
Bedrock Guardrails & Content Filtering, Prompt Injection Defense & PII Masking, KMS Encryption, IAM & VPC Isolation, Bias Detection, Fairness & Explainability, SageMaker Clarify & Model Cards, Compliance Frameworks
Provisioned Throughput & Cost Optimization, Token-Based Pricing & Model Distillation, Prompt Caching & Response Caching, CloudWatch & X-Ray Monitoring, Bedrock Invocation Logging & Metrics, Drift Detection & Operations Dashboards
Bedrock Model Evaluation, ROUGE, BLEU & BERTScore Metrics, A/B Testing & Regression Testing, Automated & Human Evaluation Pipelines, Throttling & Timeout Debugging, Model Response Quality Diagnosis
Try real exam-style questions from the free sample set. Each answer comes with a full explanation.
An insurance company is using Amazon Textract to automatically extract accident dates, damage amounts, and claimant information from scanned PDF insurance claim documents submitted by customers. When extraction accuracy is high, documents are automatically forwarded to the downstream processing pipeline, but for documents with handwriting or low image quality, extraction accuracy decreases.
Which approach MOST effectively automatically routes only low-confidence extraction results to human reviewers for verification while maintaining automatic processing for the rest and minimizing operational overhead?
Answer: B. Textract confidence score + A2I human loop
To select the correct answer, you need to identify which service natively supports confidence-based conditional routing. Amazon Textract provides a confidence score from 0 to 100 for each extracted field, and Amazon A2I (Augmented AI) can be configured as a fully managed service that uses this score as a threshold to automatically route only documents below the threshold to human reviewers via a Human Review Workflow. This combination can be set up in the console without custom coding, minimizing operational overhead.
Amazon A2I is natively integrated with AWS AI services like Textract and Rekognition, automatically triggering a Human Review Workflow based on the confidence threshold. Documents exceeding the threshold are forwarded directly to the existing downstream processing pipeline, while only documents below the threshold are delivered to reviewers through a predefined review UI. Reviewers can be selected from Amazon Mechanical Turk, a private workforce, or AWS Marketplace vendors, allowing flexible configuration to meet organizational requirements.
Option 1 (full manual review) completely abandons automatic processing, which is the opposite of the operational overhead minimization requirement. Option 3 (Comprehend sentiment analysis) is a text sentiment analysis service and is unrelated to document extraction confidence evaluation. For scenarios requiring confidence-based conditional human review, the Textract + A2I combination should be the first choice.
A development team at a financial technology startup is building a generative AI application using the Amazon Bedrock API.
Most team members are unfamiliar with the AWS SDK and Bedrock calling patterns, which is slowing development speed, and considerable time is being spent understanding complex business logic in the existing codebase. They need code understanding, explanation, and generation support within the IDE environment to boost productivity for writing Bedrock-related code, and want something immediately usable without building separate infrastructure.
Which approach is MOST appropriate?
Answer: C. Amazon CodeWhisperer
When most team members are unfamiliar with the AWS SDK and Bedrock calling patterns and spend considerable time understanding the existing codebase, a tool that immediately supports code understanding, explanation, and generation within the IDE without requiring separate infrastructure is needed. Amazon CodeWhisperer is immediately usable as a plugin for major IDEs such as VS Code and JetBrains, and directly enhances team productivity through AWS SDK and Bedrock API code completion, existing code explanation, and code generation capabilities.
Amazon CodeWhisperer is an AI coding assistant developed by AWS, now integrated into Amazon Q Developer. It supports autocomplete and full function generation for AWS-specific code patterns such as the Bedrock InvokeModel API and the boto3 library, and also provides line-by-line explanations of existing code and refactoring suggestions. A free tier is available for individual developers, and it can be used immediately just by installing the IDE plugin without building any separate servers or infrastructure.
Option 1, Bedrock Playground, is a tool for testing prompts in the web console and is not integrated into the IDE, while Option 2, AWS Cloud9, is an IDE environment itself but lacks AI code completion or code explanation features. When an exam question requires IDE integration + AWS code autocomplete + immediate use without separate infrastructure, choose Amazon CodeWhisperer or Amazon Q Developer.
A retail company is developing two features using Amazon Bedrock.
The first is a semantic search feature that analyzes customer reviews to recommend similar products. The second is a feature that generates detailed product descriptions in natural language in response to customer inquiries. The development team tried to use a single model for both features, but neither search accuracy nor generation quality met expectations.
Which approach BEST meets the requirements for both semantic search and text generation while being cost-effective?
Answer: B. Separate Titan Embeddings + Claude
Semantic search and text generation require fundamentally different model architectures. Semantic search needs an embedding model that converts text into high-dimensional vectors, and Amazon Titan Embeddings is optimized for this purpose. Text generation requires a generative model specialized in natural language understanding and generation, where Claude excels at producing detailed product descriptions. Separating the two models by purpose allows BEST-in-class performance for each task while achieving a more cost-effective architecture than a single general-purpose model.
Amazon Titan Embeddings is a dedicated embedding model that converts text into dense vectors, optimized for meaning-based retrieval through cosine similarity calculations. It integrates with vector stores such as Amazon OpenSearch Serverless and Aurora PostgreSQL as a core component of Bedrock Knowledge Bases. Claude provides superior contextual understanding and natural language generation, making it ideal for generating detailed product descriptions in response to customer inquiries.
Using a single Claude model as in Option 1 degrades embedding quality, making it difficult to achieve the required semantic search accuracy. Option 4's OpenSearch keyword search uses traditional exact-word matching and does not satisfy the semantic (meaning-based) search requirement. When an exam question asks for both semantic search and text generation, separating embedding and generative models is the AWS best practice.
A fintech company operates a loan review assistance system based on Amazon Bedrock.
The system is orchestrated by Step Functions, with Lambda functions calling Bedrock models and storing results in DynamoDB. Recently, the total processing time for some requests has increased significantly beyond expectations, but it is difficult to identify which service segment is causing the bottleneck.
Which approach MOST effectively visualizes the request flow across multiple services and precisely identifies latency segments while minimizing operational overhead?
Answer: D. AWS X-Ray distributed tracing
To select the correct answer, you first need to understand the difference between distributed tracing and log analysis. AWS X-Ray visualizes the end-to-end flow of requests across Step Functions, Lambda, DynamoDB, and Bedrock through a service map, and precisely measures latency for each service segment at the granularity of segments. This simultaneously satisfies both critical requirements of visualizing request flow across multiple services and precisely identifying latency segments. Lambda, Step Functions, and DynamoDB support native integration with X-Ray, enabling instrumentation without additional infrastructure management, thereby minimizing operational overhead.
AWS X-Ray visualizes the entire path that requests travel through each service as a graph via the Service Map, providing granular latency data broken down into Segments and Subsegments. For Bedrock invocations, subsegments can be added via the X-Ray SDK within Lambda functions to separately track model inference time. The X-Ray analytics console provides response time distribution, error rates, and anomaly detection, enabling operations teams to rapidly identify bottlenecks.
Manual analysis with CloudWatch Logs (Option 1) requires reviewing each service's logs individually, making it difficult to identify bottlenecks spanning multiple services and increasing operational overhead. CloudTrail (Option 2) is a service for security auditing and compliance purposes and does not provide functionality to measure latency per segment or trace request flow between services. When exam questions include keywords such as distributed tracing, service map, end-to-end request flow, or bottleneck identification, AWS X-Ray is the correct answer.
An e-commerce company is building a customer inquiry chatbot using Amazon Bedrock.
The development team must select the optimal foundation model by comprehensively considering response quality, latency, and cost. The team has prepared 500 real customer inquiry datasets and wants to combine automated scoring for accuracy and relevance with subjective quality evaluations from domain experts. They need to systematically compare multiple candidate models under identical conditions, manage evaluation results centrally, and minimize operational overhead.
Which approach MOST effectively meets these requirements?
Answer: B. Bedrock Model Evaluation Job
An e-commerce company needs to compare multiple foundation models using 500 real customer inquiry datasets, combining automated scoring with domain expert evaluation while minimizing operational overhead. Amazon Bedrock Model Evaluation Job supports both evaluation types—automated evaluation and human evaluation—within a single fully managed service, enabling systematic comparison of multiple models on identical datasets with centralized result management in the Bedrock console. No separate infrastructure or custom scoring code is required, minimizing operational overhead.
Amazon Bedrock Model Evaluation Job offers two evaluation types. Automated evaluation calculates quantitative metrics such as ROUGE, BERTScore, and Perplexity, while human evaluation collects subjective quality scores through Amazon Mechanical Turk or an internal expert team. Users can upload custom JSONL-format datasets to S3 or use built-in curated datasets, and evaluation results are displayed as a side-by-side score comparison table per model in the console.
Option 1, CloudWatch latency metrics comparison, only measures latency and cannot assess response quality or accuracy, making it inadequate for comprehensive model comparison. When a question requires automated evaluation + human evaluation + multi-model comparison + minimized operational overhead simultaneously, Bedrock Model Evaluation Job is the correct choice.
The AWS Certified Generative AI Developer - Professional (AIP-C01) exam consists of 75 questions with a 180-minute time limit.
The passing score for the AIP-C01 exam is 750 out of 1000.
The AIP-C01 sample set (20 questions with full explanations) is free. The full bank of 340+ questions is available with a CloudMasterIT subscription.
The AIP-C01 certification is valid for 3 years after passing. Recertification is required after that.