Free AWS Certified Generative AI Developer - Professional (AIP-C01) Practice Questions

Sample AIP-C01 Practice Questions

Browse all 10 free AWS Certified Generative AI Developer - Professional practice questions below.

1. A financial services company needs to build a document analysis application that extracts key information from loan applications. The solution must use a foundation model from Amazon Bedrock and comply with data residency requirements that mandate all data remains in the eu-central-1 region. Which approach ensures compliance while optimizing for cost?

   • Use Amazon Bedrock in eu-central-1 with the Anthropic Claude 3 Sonnet model and enable model invocation logging to an S3 bucket in the same region.
   • Deploy a self-hosted open-source LLM on EC2 instances in eu-central-1 using SageMaker endpoints for inference.
   • Implement Amazon Textract in eu-central-1 for document processing instead of using foundation models.
   • Use Amazon Bedrock in us-east-1 with cross-region replication to eu-central-1 for compliance reporting.

2. A development team is building a RAG (Retrieval-Augmented Generation) application using Amazon Bedrock Knowledge Bases with OpenSearch Serverless as the vector store. They need to index 500,000 technical documents with an average size of 50KB each. The application requires sub-second query response times and must support metadata filtering by document type, date, and department. What is the MOST effective chunking strategy for this use case?

   • Store entire documents as single chunks without splitting to preserve complete context.
   • Use sentence-level chunking with each sentence as a separate chunk to maximize retrieval precision.
   • Use fixed-size chunking with 512 tokens per chunk and 50-token overlap, storing metadata with each chunk.
   • Use semantic chunking based on paragraph boundaries with variable chunk sizes up to 2048 tokens.

3. An e-commerce company is implementing a chatbot using Amazon Bedrock with the Claude 3.5 Sonnet model. The chatbot must access real-time inventory data from a DynamoDB table and current promotional pricing from an API Gateway endpoint. The solution must minimize latency and support concurrent user sessions. Which implementation approach is MOST appropriate?

   • Configure Bedrock Agents with action groups that invoke Lambda functions to query DynamoDB and API Gateway.
   • Pre-fetch all inventory and pricing data into the prompt context for each user session using Bedrock Knowledge Bases.
   • Use prompt engineering to include DynamoDB query syntax in the system prompt and parse SQL-like queries from the model's response.
   • Implement a custom orchestration layer using Step Functions to coordinate Bedrock API calls with data retrieval from DynamoDB and API Gateway.

4. A healthcare provider is developing a clinical decision support system using Amazon Bedrock. The system must prevent the model from providing medical diagnoses, medication dosage recommendations, or contradicting established clinical guidelines. The solution must also log all blocked responses for compliance auditing. Which approach provides the MOST comprehensive protection?

   • Fine-tune the Bedrock model on examples of appropriate clinical decision support responses to teach it proper boundaries.
   • Add detailed instructions in the system prompt explicitly telling the model not to provide diagnoses or medication recommendations.
   • Implement Bedrock Guardrails with content filters, denied topics for medical diagnoses and medications, and word filters for clinical terminology, with all configurations logged to CloudWatch Logs.
   • Deploy a custom Lambda function that uses Amazon Comprehend Medical to detect medical entities in responses and blocks them before returning to users.

5. A global enterprise is deploying a multi-lingual customer service application using Amazon Bedrock. The application must support 20 languages, process 10,000 requests per minute during peak hours, and maintain sub-2-second response times. Cost optimization is a priority. Which architecture BEST meets these requirements?

   • Implement request queuing with SQS and batch process requests every 30 seconds to reduce the number of Bedrock API calls.
   • Deploy 20 separate SageMaker endpoints, one for each language, with model-specific fine-tuning to optimize language performance.
   • Use exclusively On-Demand Bedrock invocations with aggressive prompt caching to reduce token costs.
   • Use Provisioned Throughput for the base load of 5,000 RPM and On-Demand mode for burst traffic, with CloudWatch metrics triggering Auto Scaling for Provisioned Throughput adjustments.

6. A retail company is building a product recommendation system using Amazon Bedrock. The system needs to analyze customer purchase history stored in Amazon DynamoDB and product catalogs in Amazon S3. The solution must generate personalized recommendations in real-time with sub-second latency while maintaining cost efficiency. Which architecture BEST meets these requirements?

   • Use Amazon Bedrock Knowledge Bases to index product catalog from S3, implement a Lambda function to retrieve customer history from DynamoDB, and use Bedrock Agents with action groups to orchestrate the recommendation generation with Claude 3 Haiku for fast inference.
   • Load all product catalogs and customer history into the Bedrock prompt context for each request, using Claude 3 Opus for maximum recommendation quality.
   • Deploy a custom fine-tuned model on SageMaker with real-time endpoints, caching all product embeddings in ElastiCache for fast retrieval.
   • Use Amazon Personalize for recommendations instead of foundation models, as it's specifically designed for this use case.

7. A legal firm is implementing a document analysis system using Amazon Bedrock Knowledge Bases. The system must process contracts ranging from 10 pages to 500 pages, extract key clauses, and answer questions about specific terms. Documents contain tables, numbered lists, and cross-references between sections. Which chunking strategy provides the BEST balance between retrieval accuracy and context preservation?

   • Chunk documents by individual sentences to maximize retrieval precision for specific legal terms and definitions.
   • Store entire documents as single chunks to preserve complete context, using document-level embeddings for retrieval.
   • Implement hierarchical chunking that preserves document structure with parent-child relationships, using semantic chunking for paragraphs within sections while maintaining section headers as metadata for each chunk.
   • Use fixed-size chunking with 1024 tokens and 200-token overlap to ensure consistent chunk sizes across all document types.

8. A software development team is building an AI coding assistant using Amazon Bedrock. The assistant must support multiple programming languages, understand project context from a Git repository, and provide code suggestions that follow the team's coding standards stored in markdown files. Which combination of services and techniques provides the MOST effective solution? (Select TWO)

   • Store all repository code in the system prompt for each request to ensure the model has complete project context.
   • Implement a Bedrock Agent with action groups that can execute Git commands through Lambda functions to analyze repository structure, file history, and recent changes for context-aware suggestions.
   • Configure Amazon Bedrock Knowledge Bases with an S3 data source containing the Git repository code and coding standards documents, using appropriate metadata filters for language and file type.
   • Use Amazon Q Developer directly instead of building a custom solution, as it's specifically designed for code assistance with IDE integration.
   • Fine-tune a foundation model on the team's codebase to create a specialized code generation model that inherently understands their patterns.

9. A healthcare research organization is using Amazon Bedrock to analyze medical literature and generate research summaries. The application must ensure that generated content is grounded in the source documents and does not include fabricated citations or statistics. The solution must also provide traceability to original sources. Which approach provides the MOST reliable grounding and citation accuracy?

   • Use Claude 3 Opus with temperature set to 0 to ensure deterministic, factual outputs that don't include creative elaborations.
   • Fine-tune the foundation model on the organization's medical literature corpus to ensure it only generates content about known sources.
   • Implement Bedrock Knowledge Bases with contextual grounding checks enabled in Guardrails, configure citation extraction in the retrieval response, and use a post-processing Lambda function to verify that all cited sources exist in the retrieved context before returning results.
   • Add explicit instructions in the system prompt telling the model to only cite sources from the provided context and never fabricate references.

10. A financial services company is deploying a customer-facing chatbot using Amazon Bedrock. The chatbot handles account inquiries and must be protected against prompt injection attacks that could trick it into revealing other customers' information or performing unauthorized actions. Which combination of security controls provides the MOST comprehensive protection? (Select TWO)

    • Design the agent architecture to separate user context from system instructions using Lambda pre-processing, validate all user inputs against allow-lists, and implement IAM-scoped tool permissions that limit what actions the chatbot can perform regardless of instructions.
    • Implement rate limiting on the chatbot API to prevent attackers from attempting multiple injection variations.
    • Use a smaller, less capable model that is less likely to follow complex injection instructions.
    • Store all customer data in encrypted format so that even if the model is manipulated, it cannot reveal readable information.
    • Implement Bedrock Guardrails with content filters for harmful content, denied topics for account manipulation attempts, and custom word filters for prompt injection patterns.