View Full Microsoft AI-102 Exam Dumps and Practice Test Dumps.

 

Question 21

Which capability allows a language model to use external information when generating a response?

1. Grounding
2. Image cropping
3. Speech synthesis
4. Video encoding

Correct Answer: 1

Explanation

Grounding provides a generative AI model with relevant external information that can be used when producing a response. This is especially useful when an application needs answers based on current, private, or organization-specific information that may not be contained in the model’s original training data. Grounding can be implemented through retrieval-based approaches, where relevant content is retrieved and supplied as context. Image cropping, speech synthesis, and video encoding do not provide external textual context to a language model. Grounding therefore helps improve factual relevance and alignment with application-specific information.

Question 22

Which Azure AI Language capability can identify the main topic discussed in a document?

1. Speech translation
2. Key phrase extraction
3. Text-to-speech
4. Object detection

Correct Answer: 2

Explanation

Key phrase extraction identifies important concepts or phrases within unstructured text. It can help applications understand the major topics discussed in documents without requiring a complete manual review. For example, a customer-feedback system could extract phrases related to delivery delays, product quality, or customer support. Speech translation processes spoken language, text-to-speech generates audio, and object detection analyzes visual content. Key phrase extraction is therefore useful when an application needs a concise representation of the important topics contained in text. It can also support search, categorization, and content-analysis workflows.

Question 23

An application must determine whether a submitted document is written in English, French, or German. Which capability should be used?

1. Entity recognition
2. Language detection
3. Sentiment analysis
4. Key phrase extraction

Correct Answer: 2

Explanation

Language detection identifies the language used in a piece of text. This can be useful when applications receive multilingual content and need to select an appropriate processing workflow. For example, an application could detect whether incoming text is English, French, or German and then route it to language-specific processing or translation. Entity recognition identifies named entities, sentiment analysis evaluates emotional tone, and key phrase extraction identifies important concepts. Since the requirement is specifically to determine which language a document uses, language detection is the appropriate Azure AI Language capability for the solution.

Question 24

Which approach can help prevent a generative AI application from returning unsupported information?

1. Ground responses with trusted data
2. Increase image resolution
3. Disable logging
4. Remove all instructions

Correct Answer: 1

Explanation

Grounding generative AI responses with trusted and relevant data can reduce the likelihood of unsupported information being presented as factual. In a retrieval-augmented solution, the application retrieves appropriate content from approved sources and provides that content as context to the model. Prompt instructions can further tell the model to base its response on the supplied information and indicate when sufficient information is unavailable. Increasing image resolution, disabling logging, or removing instructions does not address unsupported textual responses. Grounding should also be combined with evaluation and monitoring to assess response quality.

Question 25

Which feature is most useful when an AI application must return responses in a predictable structured format?

1. Prompt instructions
2. Image classification
3. Speech recognition
4. OCR

Correct Answer: 1

Explanation

Clear prompt instructions can guide a generative AI model toward a predictable response structure. A developer can specify required fields, formatting rules, ordering, and other output constraints in the prompt. For example, an application could request a response containing a summary, priority, and recommended action in a defined structure. Although structured output features may also be available depending on the model and service, prompt design remains an important part of controlling model behavior. Image classification, speech recognition, and OCR solve different problems and do not directly control the format of generated textual responses.

Question 26

A developer wants to use an existing language model through an application. What is typically required before the application can send requests to that model?

1. Model deployment
2. Image labeling
3. OCR extraction
4. Speech recording

Correct Answer: 1

Explanation

A model generally needs to be deployed in the appropriate AI environment before an application can send requests to it. Deployment provides an endpoint or connection through which supported application calls can be made and can also establish configuration such as the selected model and deployment settings. Image labeling, OCR extraction, and speech recording are unrelated to making a language-model deployment available. Developers should also consider authentication, permissions, supported APIs, quotas, and model availability when integrating the deployed model into an application. Proper deployment configuration is an important step between model selection and application usage.

Question 27

Which capability can identify the emotional tone of a customer support message?

1. Language detection
2. Sentiment analysis
3. OCR
4. Image segmentation

Correct Answer: 2

Explanation

Sentiment analysis evaluates text to determine its expressed sentiment, commonly represented through categories such as positive, negative, or neutral, together with confidence information. A customer-support application can use sentiment analysis to identify messages that may indicate dissatisfaction or positive feedback and route them appropriately. Language detection determines the language of text, OCR extracts text from images, and image segmentation works with visual content. Sentiment analysis should not be treated as a perfect measurement of a person’s actual emotions; it evaluates the content of the submitted text. Its results should therefore be interpreted according to the application’s context.

Question 28

An organization wants to summarize lengthy customer comments automatically. Which capability is most directly applicable?

1. Text summarization
2. Object detection
3. Speech recognition
4. Image tagging

Correct Answer: 1

Explanation

Text summarization generates a shorter representation of longer textual content while attempting to preserve the important information. It can be useful for customer feedback, reports, case notes, and other documents where users need to understand the main points quickly. An application can process lengthy comments and produce concise summaries that support faster review. Object detection and image tagging analyze visual content, while speech recognition converts spoken audio into text. Text summarization is therefore the capability that most directly addresses the requirement to reduce lengthy customer comments into useful, concise information.

Question 29

Which service is designed to analyze the content and characteristics of images?

1. Azure AI Vision
2. Azure AI Language
3. Azure Speech
4. Azure AI Search

Correct Answer: 1

Explanation

Azure AI Vision provides computer vision capabilities for analyzing images and visual content. Depending on the supported feature, applications can use it for tasks such as image analysis, OCR, object detection, image tagging, and other visual scenarios. Azure AI Language focuses on natural language processing, Azure Speech handles speech and audio scenarios, and Azure AI Search provides search and knowledge-mining capabilities. When an application needs to understand information contained within images, Azure AI Vision is therefore a relevant service. The specific Vision feature should then be selected according to the required image-analysis task.

Question 30

What is the purpose of a content filter in a generative AI application?

1. Increase storage
2. Detect potentially harmful content
3. Create database schemas
4. Convert audio to text

Correct Answer: 2

Explanation

Content filters help identify and manage potentially harmful or inappropriate content in generative AI interactions. Depending on the service and configuration, filtering can be applied to prompts, generated responses, or both. This provides an additional safety layer for applications that interact with users and can help reduce exposure to categories of content that the application should not generate or process. Content filtering does not increase storage capacity, create database schemas, or convert audio into text. Developers should combine content filtering with responsible AI practices, application controls, testing, and monitoring rather than treating filtering as the only safety mechanism.

Question 31

A developer needs to identify text printed on a street sign within an image. Which capability should be considered?

1. OCR
2. Sentiment analysis
3. Entity linking
4. Text summarization

Correct Answer: 1

Explanation

Optical character recognition, or OCR, is designed to extract readable text from images. A street sign captured by a camera can be processed using OCR so that the printed characters become machine-readable text for further application processing. Once extracted, the text could potentially be translated, searched, classified, or analyzed using other services. Sentiment analysis evaluates emotional tone, entity linking connects recognized entities to knowledge sources, and text summarization creates concise representations of existing text. Because the immediate requirement is to recognize words appearing visually on the sign, OCR is the appropriate capability.

Question 32

Which Azure AI Speech capability converts spoken words into another language?

1. Image analysis
2. Speech translation
3. Document extraction
4. Semantic ranking

Correct Answer: 2

Explanation

Speech translation processes spoken language and produces translated speech or text according to the supported configuration. It can be used in scenarios such as multilingual conversations, meetings, customer interactions, and applications that need to bridge language differences. Image analysis works with visual information, document extraction focuses on structured information in documents, and semantic ranking improves search relevance. Speech translation can involve speech recognition as part of the overall process before or alongside translation. Therefore, when the requirement specifically involves converting spoken content from one language into another, the appropriate capability is speech translation.

Question 33

Why might an engineer use a vector index in a generative AI application?

1. To store audio recordings only
2. To support similarity-based retrieval
3. To replace authentication
4. To generate speech

Correct Answer: 2

Explanation

A vector index stores numerical representations of content that can be compared for semantic similarity. In generative AI applications, this can support retrieval of documents or passages that are conceptually related to a user’s query even when they do not share the same exact keywords. The retrieved information can then be supplied to a model as grounding context. Vector indexes do not replace authentication, generate speech, or function only as audio storage. When implementing a retrieval-augmented generation architecture, vector search can therefore help identify relevant information for generating context-aware responses.

Question 34

Which practice helps improve the reliability of a generative AI application before release?

1. Skip testing
2. Evaluate representative scenarios
3. Remove monitoring
4. Ignore failed responses

Correct Answer: 2

Explanation

Evaluating representative scenarios helps determine whether a generative AI application behaves as expected under realistic conditions. Test cases should reflect the types of prompts, data, users, and edge cases that the application is expected to encounter. Evaluation can examine factors such as relevance, groundedness, correctness, safety, and consistency depending on the solution. Skipping testing or ignoring failed responses can leave important problems undiscovered, while removing monitoring reduces visibility after deployment. Representative evaluation provides evidence for refining prompts, retrieval settings, model choices, and application controls before the system is made available to users.

Question 35

A company wants to search a large collection of documents and return relevant passages for an AI application. Which service is appropriate?

1. Azure AI Search
2. Azure Speech
3. Azure AI Vision
4. Azure AI Language only

Correct Answer: 1

Explanation

Azure AI Search provides capabilities for indexing and searching large collections of content. It can support keyword search, vector search, hybrid search, semantic capabilities, and other features depending on the configured solution. This makes it useful for retrieving relevant documents or passages that can then be supplied to a generative AI model as grounding information. Azure Speech is intended for speech workloads, Azure AI Vision focuses on visual content, and Azure AI Language provides language-processing capabilities. For an application that needs document retrieval as part of an AI workflow, Azure AI Search is a suitable service.

Question 36

Which action can help protect an AI application from unauthorized access to its Azure resources?

1. Use appropriate authentication and authorization
2. Disable identity controls
3. Publish credentials in prompts
4. Share secrets with users

Correct Answer: 1

Explanation

Authentication and authorization controls help ensure that only permitted identities can access AI resources and perform approved operations. An application should use appropriate Microsoft identity and access mechanisms and follow least-privilege principles when granting permissions. Credentials and secrets should never be exposed in prompts or shared unnecessarily with users. Disabling identity controls increases the risk of unauthorized access rather than reducing it. Security should be considered throughout the AI application’s lifecycle, including development, deployment, and operation. Proper identity configuration is therefore an important component of protecting AI resources and connected data.

Question 37

A developer wants to create a conversational application that maintains relevant information from earlier interactions. What should be considered?

1. Conversation context
2. Image resolution
3. OCR confidence only
4. Network cables

Correct Answer: 1

Explanation

Conversation context allows an AI application to use relevant information from earlier interactions when responding to subsequent user messages. Maintaining appropriate context can help a conversational system understand references, follow multi-step discussions, and provide more coherent responses. Developers must decide what information should be retained, how much context should be supplied to the model, and how sensitive information should be handled. Image resolution, OCR confidence, and physical networking are unrelated to maintaining conversational state. Context management should also account for token limits, privacy requirements, and the application’s intended user experience.

Question 38

Which technique can improve retrieval when users and documents use different but related words?

1. Vector search
2. Audio compression
3. Image cropping
4. Speech synthesis

Correct Answer: 1

Explanation

Vector search can identify content based on semantic similarity rather than requiring exact keyword matches. Text is represented as vectors, and queries can be compared with indexed content to find conceptually related information. This can improve retrieval when users phrase a question differently from the wording used in the stored documents. For example, a user might ask about automobile maintenance while a document discusses vehicle servicing. Audio compression, image cropping, and speech synthesis do not provide semantic document retrieval. Vector search is particularly useful in generative AI architectures where retrieved content is used to ground model responses.

Question 39

A team needs to determine whether an AI model’s responses meet predefined quality requirements. What should they perform?

1. Model evaluation
2. Resource deletion
3. Image resizing
4. Credential sharing

Correct Answer: 1

Explanation

Model evaluation measures how well an AI model or AI-powered application performs against defined criteria. Evaluation criteria should reflect the intended use case and may include response relevance, accuracy, groundedness, safety, coherence, or other measurable qualities. A team can use representative test data and scenarios to identify weaknesses and compare different configurations. Resource deletion, image resizing, and credential sharing do not establish whether model responses meet quality requirements. Evaluation is especially important before production deployment and should continue as the application, models, prompts, data, or user requirements change.

Question 40

What is an important consideration when integrating an AI model into a production application?

1. Ignore service limits
2. Consider latency, cost, and scalability
3. Remove authentication
4. Avoid monitoring

Correct Answer: 2

Explanation

Production AI applications must consider operational factors such as latency, cost, scalability, throughput, quotas, reliability, and expected user demand. A model that performs well in a small development test may not provide the required performance or cost profile when usage increases. Engineers should evaluate expected workload characteristics and configure the solution accordingly. Authentication and monitoring should also remain in place to support security and operational visibility. Ignoring service limits or avoiding monitoring can create deployment problems. Therefore, latency, cost, and scalability are important considerations when integrating an AI model into a production application.