View Full Amazon AWS Certified Generative AI Developer – Professional AIP-C01 Exam Dumps and Practice Test Dumps

 

Question 141

Which technique helps a model follow a required response format by providing examples in the prompt?

1. Zero-shot prompting
2. Few-shot prompting
3. Vector indexing
4. Model monitoring

Correct Answer: 2

Explanation

Few-shot prompting provides the model with examples of desired inputs and outputs before asking it to process a new request. These examples help establish the expected format, style, reasoning pattern, or classification behavior. For instance, a developer can provide several examples showing how customer requests should be categorized before presenting a new request. Zero-shot prompting provides no examples. Vector indexing is associated with retrieval systems, while model monitoring focuses on observing application behavior rather than directly teaching response patterns through examples.

Question 142

A company wants its AI assistant to retrieve relevant information from millions of internal documents. Which component is most important for efficient semantic retrieval?

1. Vector index
2. API Gateway
3. CloudTrail trail
4. IAM policy

Correct Answer: 1

Explanation

A vector index enables efficient similarity searches across embeddings representing documents or other data. In a RAG architecture, documents are converted into embeddings and stored in a vector-capable search system. When a user submits a query, the query can also be converted into an embedding and compared against stored vectors to identify semantically similar content. This approach scales better than relying only on exact keyword matching for many natural-language use cases. API Gateway, CloudTrail, and IAM provide API management, auditing, and access control rather than vector retrieval.

Question 143

A developer wants to improve RAG results when keyword search finds relevant terms but misses documents that use different terminology. Which retrieval strategy should be considered?

1. Semantic retrieval
2. Disabling embeddings
3. Increasing temperature
4. Reducing the document collection

Correct Answer: 1

Explanation

Semantic retrieval uses embeddings to identify content based on meaning rather than requiring exact keyword matches. It is particularly useful when users and documents use different terminology to describe the same concept. For example, a query mentioning “vehicle coverage” may retrieve content discussing “automobile insurance” even when the exact phrase is not present. Hybrid retrieval can also combine semantic and keyword approaches when both meaning and exact terms are important. Temperature affects generation rather than retrieval, while reducing the document collection may remove useful information.

Question 144

A developer is designing a RAG pipeline and wants to prevent retrieved documents from exceeding the model’s context window. What should the developer optimize?

1. Chunking and context selection
2. Model temperature
3. IAM policies
4. CloudTrail retention

Correct Answer: 1

Explanation

Chunking and context selection help control how much information is passed to the foundation model. Large source documents can be divided into meaningful chunks, and the retrieval system can return only the most relevant chunks for a particular query. Developers should consider chunk size, overlap, metadata, retrieval count, and reranking when designing the pipeline. Sending excessive context can increase cost and latency and may exceed model context limits. Temperature, IAM policies, and CloudTrail retention do not directly control the amount of retrieved information sent to the model.

Question 145

Which technique can help a RAG system rank retrieved documents according to their relevance before sending them to the model?

1. Reranking
2. Encryption
3. Token rotation
4. API throttling

Correct Answer: 1

Explanation

Reranking evaluates an initial set of retrieved documents and orders them according to their relevance to the user’s query. This can improve RAG quality when the first-stage retrieval returns several potentially relevant documents but their ranking is imperfect. A reranker can help select the most useful context before it is provided to the foundation model. Encryption protects data, token rotation manages credentials or secrets, and API throttling controls request rates. These capabilities address different concerns and do not directly improve document relevance ranking.

Question 146

A company wants to customize a foundation model so that it consistently follows a specialized organizational style and performs better on a specific task. Which approach should the developer evaluate?

1. Fine-tuning or model customization
2. Increasing API timeout
3. Adding CloudTrail logs
4. Increasing S3 storage capacity

Correct Answer: 1

Explanation

Fine-tuning or other supported model-customization approaches can adapt a foundation model to specialized tasks, patterns, or organizational requirements. This can be useful when prompting alone does not provide sufficiently consistent behavior. The training data should be carefully prepared, representative, and appropriate for the intended task. Developers should also evaluate whether customization is actually necessary because prompt engineering or RAG may be more suitable for some requirements. API timeouts, CloudTrail logging, and S3 capacity do not change the model’s learned behavior.

Question 147

A developer is building an AI application where users submit requests through an API, and each request may trigger several independent processing tasks. Which design can improve scalability?

1. Sequential processing inside one long-running request
2. Event-driven parallel processing
3. Storing requests only in application logs
4. Increasing prompt temperature

Correct Answer: 2

Explanation

Event-driven parallel processing can improve scalability when multiple independent tasks can execute separately. Events can trigger appropriate workers or serverless functions, allowing workloads to scale based on demand. Independent tasks can potentially execute concurrently instead of forcing every operation into one long-running request. This can reduce latency and improve resource utilization. A single synchronous request may become a bottleneck as workload increases. Application logs are not designed to function as a workflow mechanism, and temperature affects model output rather than application scalability.

Question 148

An enterprise wants its AI application to automatically invoke a tool, inspect the result, and decide whether another tool is needed. Which architecture best supports this behavior?

1. Agentic workflow
2. Static HTML application
3. Object storage only
4. Basic keyword search

Correct Answer: 1

Explanation

An agentic workflow allows an AI system to reason about a task, select available tools, execute actions, inspect results, and continue toward a desired outcome. This differs from a fixed workflow where every step is predetermined. Agentic systems can be useful for tasks such as researching information, calling APIs, retrieving documents, and performing multi-step business operations. Tool definitions, authorization, input validation, and safeguards remain important because the model is involved in deciding which actions to perform. Static websites, storage, and keyword search do not provide this orchestration behavior.

Question 149

A developer wants an AI agent to use tools safely. Which control is most important for preventing malformed or dangerous tool parameters from reaching a backend service?

1. Input validation
2. Higher temperature
3. Larger embeddings
4. Longer prompts

Correct Answer: 1

Explanation

Input validation checks tool parameters before they are passed to backend systems. This can prevent invalid values, unexpected formats, unauthorized identifiers, or potentially dangerous inputs from reaching sensitive services. Validation should be performed by trusted application logic rather than relying solely on the model to generate safe parameters. Developers can also combine validation with authorization, rate limits, and backend controls. Increasing temperature, changing embedding dimensions, or adding more prompt text does not provide reliable protection against malformed or unsafe tool inputs.

Question 150

A company wants to expose a generative AI service to many clients while preventing excessive requests from a single client from overwhelming the backend. Which API control should be considered?

1. API throttling
2. Prompt expansion
3. Higher embedding dimensions
4. Increased model temperature

Correct Answer: 1

Explanation

API throttling limits the rate at which clients can send requests to an application. This helps protect backend services from excessive traffic and can improve overall reliability and resource availability. Amazon API Gateway provides throttling capabilities that can be used as part of an API management strategy. Appropriate limits should be based on expected workload, client requirements, and backend capacity. Prompt size and temperature affect model behavior, while embedding dimensions affect vector representations. None of these directly controls how quickly clients can submit API requests.

Question 151

Which AWS service can collect metrics and logs to help monitor a generative AI application’s operational behavior?

1. AWS KMS
2. Amazon CloudWatch
3. Amazon S3 Glacier
4. AWS Secrets Manager

Correct Answer: 2

Explanation

Amazon CloudWatch provides monitoring capabilities for AWS resources and applications through metrics, logs, alarms, and related observability features. A generative AI application can use CloudWatch to monitor indicators such as invocation counts, latency, errors, resource utilization, and application logs. These signals can help developers identify performance issues and operational failures. AWS KMS manages encryption keys, S3 Glacier is intended for archival storage, and Secrets Manager stores sensitive credentials. CloudWatch is therefore the most appropriate service for operational monitoring.

Question 152

A generative AI application experiences a sudden increase in model invocation failures. The developer wants to determine whether the problem is caused by application changes or the underlying AWS service interaction. What should the developer examine first?

1. Logs, metrics, and invocation errors
2. The model’s temperature only
3. The application’s website design
4. The number of S3 folders

Correct Answer: 1

Explanation

Logs, metrics, and detailed invocation errors provide evidence about where and why model requests are failing. Developers can examine request patterns, error types, latency, throttling indicators, authentication failures, input validation issues, and recent deployment changes. This evidence helps distinguish application-level problems from service configuration or capacity-related issues. Changing temperature does not address invocation failures, while website design and S3 folder counts are generally unrelated. Effective troubleshooting should begin with observable evidence before configuration changes are made.

Question 153

A company wants to test a new model version with a small percentage of production traffic before moving all users to it. Which deployment strategy is most appropriate?

1. Canary deployment
2. Immediate full replacement
3. Permanent rollback
4. Manual prompt deletion

Correct Answer: 1

Explanation

A canary deployment introduces a new version to a small portion of traffic while the existing version continues serving most users. Developers can monitor quality, latency, errors, safety, and cost before gradually increasing traffic to the new version. If problems occur, traffic can be redirected to the stable version with limited user impact. Immediate full replacement increases deployment risk because every user is affected at once. Canary testing is therefore useful for validating generative AI application or model changes under controlled production conditions.

Question 154

A developer needs to compare two foundation models for an AI summarization workload. Which evaluation approach provides the strongest comparison?

1. Use representative inputs and predefined quality metrics
2. Select the model with the longest name
3. Compare only the models’ temperatures
4. Test each model with unrelated prompts

Correct Answer: 1

Explanation

A meaningful model comparison requires representative inputs and predefined evaluation criteria. The same test dataset should be used with both models so that results can be compared consistently. Metrics may include factual accuracy, relevance, groundedness, response quality, latency, cost, and safety depending on the application’s requirements. Comparing only configuration parameters does not establish which model performs better for the intended workload. Unrelated prompts can also produce misleading conclusions. Evaluation should therefore reflect actual production use cases and measurable business or technical requirements.

Question 155

A developer wants to reduce the number of input tokens sent to a foundation model without significantly reducing the information needed to answer a user’s question. Which approach is most appropriate?

1. Remove irrelevant context
2. Add more documents
3. Increase response temperature
4. Duplicate system instructions

Correct Answer: 1

Explanation

Removing irrelevant context can reduce input token usage while preserving the information that is actually required to answer the user’s question. In RAG systems, developers can improve retrieval precision, reduce unnecessary chunks, use metadata filtering, and select only the most relevant context. This can lower inference cost and latency while potentially improving answer quality by reducing noise. Adding documents or duplicating instructions increases the prompt size. Temperature affects output variability and does not directly reduce the number of input tokens.

Question 156

An AI application must ensure that only authenticated users can access a protected API endpoint. Which capability should be included in the API architecture?

1. Authentication and authorization
2. Higher model temperature
3. Larger vector indexes
4. Additional prompt examples

Correct Answer: 1

Explanation

Authentication verifies the identity of the caller, while authorization determines whether that identity is permitted to perform a particular operation. Both controls are important when protecting an AI application’s API. Depending on the architecture, API Gateway and supporting identity services can enforce appropriate access controls before requests reach backend application logic. Prompt engineering and vector indexes do not provide reliable API security. A secure design should also consider least-privilege IAM permissions, encryption, input validation, logging, and rate limiting.

Question 157

A company processes confidential customer information through a generative AI application. The security team wants to minimize unnecessary exposure of sensitive data in prompts. What should the developer consider?

1. Data minimization and sensitive-data controls
2. Increasing temperature
3. Adding duplicate customer records
4. Disabling access controls

Correct Answer: 1

Explanation

Data minimization means sending only the information required for the task instead of unnecessarily including complete sensitive records. Developers should identify sensitive fields, remove or mask information when possible, restrict access, and apply appropriate data-handling controls throughout the application. This reduces the amount of confidential information exposed to model processing and downstream systems. Increasing temperature has no privacy benefit, while duplicating sensitive records increases exposure. Disabling access controls creates additional risk. Privacy should therefore be considered throughout the complete AI application architecture.

Question 158

A developer wants to reduce hallucinations caused by retrieving too many loosely related documents. Which change is most likely to help?

1. Improve retrieval precision and limit context to relevant results
2. Increase the temperature
3. Remove all document metadata
4. Retrieve every available document

Correct Answer: 1

Explanation

Retrieving excessive or weakly related documents can introduce conflicting information and distract the model from the most relevant evidence. Improving retrieval precision and limiting the final context to high-quality results can reduce this problem. Developers can use better chunking, metadata filters, semantic retrieval, hybrid search, reranking, and appropriate retrieval limits. Increasing temperature generally does not improve factual grounding. Retrieving every available document increases noise, cost, and latency. The goal should be to provide the model with focused, relevant, and trustworthy context.

Question 159

Which capability is most useful for identifying unusual API activity and supporting an audit of AWS resource access?

1. AWS CloudTrail
2. Amazon S3 lifecycle rules
3. Amazon Bedrock Knowledge Bases
4. Amazon DynamoDB Streams

Correct Answer: 1

Explanation

AWS CloudTrail records API activity across supported AWS services and provides information useful for security analysis, governance, and auditing. Organizations can examine events to identify actions performed by users, roles, or services and investigate unusual access patterns. CloudTrail can therefore help establish an audit trail for AI applications using AWS resources. S3 lifecycle rules manage object storage transitions, Knowledge Bases support retrieval, and DynamoDB Streams capture database item changes. These services serve different purposes and do not provide the same AWS API auditing capability.

Question 160

A generative AI application has acceptable response quality but its operating cost is significantly higher than expected. The application uses a large model for simple requests. What should the developer evaluate first?

1. A smaller model that meets the quality requirements
2. A higher temperature
3. More retrieved documents
4. Larger prompts

Correct Answer: 1

Explanation

Model selection has a major effect on generative AI operating costs. If a smaller foundation model can satisfy the application’s quality, safety, and latency requirements, using it for appropriate workloads can significantly reduce inference costs. Developers should evaluate models using representative test data rather than selecting solely on model size. More retrieved documents and larger prompts generally increase token usage and cost. Temperature does not directly reduce inference pricing. A model-routing strategy can also be considered when simple and complex requests have different requirements.