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

 

Question 61

Which Azure AI service is designed to analyze spoken language and return transcribed text?

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

Correct Answer: 2

Explanation

Azure Speech provides speech-to-text capabilities that convert spoken audio into written text. This capability is useful for applications such as voice assistants, meeting transcription, call analysis, accessibility solutions, and voice-controlled applications. The service can process supported audio input and return recognized text that can then be used by other components of an AI solution. Azure AI Search focuses on search and retrieval, Azure AI Vision handles visual content, and Azure AI Language processes text. Therefore, Azure Speech is the appropriate service when an application needs to transform spoken language into machine-readable text.

Question 62

Which feature can help an AI application identify relationships between recognized entities in text?

1. Entity linking
2. Image classification
3. Speech synthesis
4. Object detection

Correct Answer: 1

Explanation

Entity linking connects recognized entities in text to information in a knowledge base or provides additional identification information for those entities. For example, when a document mentions a well-known company or location, entity linking can help associate that mention with a specific entity rather than treating it as ordinary text. This can support search, content analysis, and knowledge-mining scenarios. Image classification and object detection process visual information, while speech synthesis produces spoken output. Entity linking is therefore useful when an application needs to enrich recognized textual entities with additional contextual information.

Question 63

A developer wants to analyze a video to determine when specific objects appear. Which approach is appropriate?

1. Video analysis
2. Text summarization
3. Sentiment analysis
4. Language detection

Correct Answer: 1

Explanation

Video analysis can process visual content across video frames to identify objects, activities, or other supported characteristics. Unlike a single-image analysis operation, video scenarios require consideration of information that changes over time. An application can analyze frames or use appropriate video-processing capabilities to determine when relevant visual elements appear. Text summarization and sentiment analysis focus on language, while language detection determines the language of text. Therefore, video analysis is the appropriate approach when the requirement involves identifying visual content throughout a video rather than analyzing a single static image.

Question 64

What is the main purpose of a custom Azure AI Document Intelligence model?

1. Process specialized document layouts
2. Translate spoken conversations
3. Detect image objects
4. Generate synthetic speech

Correct Answer: 1

Explanation

A custom Document Intelligence model can be developed for document types that have specialized layouts or fields not adequately handled by suitable prebuilt models. For example, an organization may receive a proprietary form containing specific fields that must be extracted consistently. The custom model can be trained or configured using representative documents and then used to process similar documents. Speech translation, object detection, and speech synthesis address different AI workloads. A custom document model is therefore valuable when the organization needs structured extraction from documents whose layout or field requirements are specific to its business process.

Question 65

Which capability is useful for determining whether two documents discuss similar topics?

1. Semantic similarity
2. Speech recognition
3. OCR
4. Face detection

Correct Answer: 1

Explanation

Semantic similarity evaluates how closely pieces of content are related in meaning. In AI solutions, embeddings can represent documents or text passages as vectors, allowing an application to compare their semantic relationships. This can support scenarios such as document clustering, duplicate detection, recommendation systems, and semantic retrieval. Speech recognition converts audio to text, OCR extracts text from images, and face detection identifies faces in visual content. Semantic similarity does not necessarily require the documents to use identical words; documents can be related even when their wording differs. This makes semantic comparison valuable for large collections of unstructured information.

Question 66

An application must answer a user’s question using a company’s indexed documents. What should happen before generation?

1. Retrieve relevant documents
2. Delete the index
3. Disable search
4. Remove application context

Correct Answer: 1

Explanation

Relevant documents should be retrieved before generation when a generative AI application is designed to answer questions using indexed organizational content. The retrieval stage identifies information related to the user’s query and provides that information as context to the language model. This architecture helps ground the generated response in the organization’s available data instead of relying exclusively on the model’s internal knowledge. Deleting the index, disabling search, or removing context would prevent the application from obtaining the information it needs. Retrieval quality should also be evaluated because poor search results can negatively affect the generated response.

Question 67

Which Azure AI Search feature can improve ranking based on the meaning of search content?

1. Semantic ranker
2. Speech recognizer
3. OCR engine
4. Image classifier

Correct Answer: 1

Explanation

The semantic ranker in Azure AI Search can improve the ordering of search results by applying language understanding to the retrieved content. Traditional keyword retrieval can first identify candidate documents, after which semantic ranking can help determine which candidates are more relevant to the meaning of the user’s query. This can improve search experiences when exact keyword matching alone does not adequately represent relevance. Speech recognition, OCR, and image classification serve different purposes. When an application needs more meaningful ranking of search results, semantic ranking is therefore an important Azure AI Search capability to consider.

Question 68

Which approach is most appropriate for extracting a specific field from a recurring custom form?

1. Custom Document Intelligence model
2. Speech translation
3. Image captioning
4. Sentiment analysis

Correct Answer: 1

Explanation

A custom Document Intelligence model is appropriate when an organization repeatedly processes documents with a specialized layout and needs specific fields extracted. The model can be trained or configured using representative examples so that it learns the structure and information relevant to the organization’s documents. This can be useful for proprietary forms, internal applications, and specialized business paperwork. Speech translation handles spoken language, image captioning describes visual content, and sentiment analysis evaluates textual sentiment. Therefore, a custom Document Intelligence model is the appropriate choice when reliable extraction of recurring organization-specific fields is required.

Question 69

What should an application do when a retrieved knowledge source does not contain enough information to answer a question?

1. Indicate that the information is unavailable
2. Invent a detailed answer
3. Ignore the knowledge source
4. Return unrelated content

Correct Answer: 1

Explanation

When the available knowledge source does not contain enough information, an application should avoid presenting unsupported information as fact. The prompt and application logic can instruct the model to acknowledge when the retrieved context is insufficient and, where appropriate, ask the user for clarification or direct them to another source. This behavior helps reduce hallucination risk and improves transparency. Inventing an answer or returning unrelated content can mislead users. A well-designed grounded AI application should therefore establish clear behavior for situations where retrieval returns insufficient or low-confidence information.

Question 70

Which capability can extract structured fields such as names, dates, and addresses from supported documents?

1. Document Intelligence
2. Speech
3. Translator
4. Search semantic ranker

Correct Answer: 1

Explanation

Azure AI Document Intelligence can extract structured fields from supported document types. Depending on the model, extracted information can include names, dates, addresses, totals, identifiers, and other document-specific fields. This allows applications to transform unstructured or semi-structured business documents into information that can be processed by downstream systems. Azure Speech handles speech workloads, Translator converts text between supported languages, and semantic ranking improves search relevance. Document Intelligence is therefore the appropriate service when the primary requirement is extracting structured information from documents for automation, analysis, or integration with business applications.

Question 71

A developer needs to create a custom text classification solution for support tickets. What should be considered?

1. Training data and labels
2. Image resolution
3. Audio bitrate only
4. DNS configuration

Correct Answer: 1

Explanation

A custom text classification solution depends on representative training data and appropriate labels. Support tickets should be collected and categorized according to the classes the application needs to recognize, such as billing, technical support, account access, or delivery issues. The quality and coverage of the training data can strongly influence model performance. Developers should also evaluate the resulting model using representative test data and monitor its behavior after deployment. Image resolution, audio bitrate, and DNS configuration do not determine the quality of a text classification model. Good labeled data is therefore a fundamental requirement.

Question 72

What does text classification allow an AI application to do?

1. Assign text to predefined categories
2. Detect physical faces
3. Generate audio from text
4. Resize photographs

Correct Answer: 1

Explanation

Text classification assigns textual content to predefined categories based on the characteristics of the text. For example, an organization could classify customer messages into billing, complaints, technical support, or general inquiries. Classification can be implemented using suitable language-processing capabilities and may use either predefined or custom models depending on the scenario. Face detection analyzes images, text-to-speech converts written content into audio, and image resizing changes visual dimensions. Text classification is useful when an application needs to automatically route, organize, or prioritize large volumes of textual information according to known categories.

Question 73

Which factor should be considered when choosing between a prebuilt and custom AI model?

1. Document or data characteristics
2. Monitor size
3. Keyboard layout
4. Desktop wallpaper

Correct Answer: 1

Explanation

The characteristics of the organization’s data or documents should guide the choice between prebuilt and custom AI models. A prebuilt model may be appropriate when the workload matches a supported scenario and expected output fields. A custom model may be more suitable when the organization uses specialized document layouts, unique categories, or requirements that are not adequately handled by available prebuilt models. The decision should also consider development effort, data availability, expected accuracy, maintenance, and testing requirements. Monitor size, keyboard layout, and desktop appearance have no meaningful role in selecting an AI model.

Question 74

Which technique can help an AI application retrieve content using the conceptual meaning of a query?

1. Vector search
2. Exact filename matching only
3. Audio encoding
4. Image compression

Correct Answer: 1

Explanation

Vector search uses numerical representations of content to identify information that is semantically similar to a query. Instead of requiring the query and document to contain exactly the same words, vector-based retrieval can identify content with related meanings. This makes it useful for knowledge bases, document search, and retrieval-augmented generation. Exact filename matching alone cannot capture semantic relationships, while audio encoding and image compression solve unrelated technical problems. When implementing vector search, developers should also consider embedding models, indexing configuration, query behavior, relevance evaluation, and the characteristics of the source data.

Question 75

A company wants to allow users to ask questions about internal policies stored as PDF files. Which architecture is appropriate?

1. RAG with indexed documents
2. Image classification only
3. Speech synthesis only
4. Static HTML without retrieval

Correct Answer: 1

Explanation

A retrieval-augmented generation architecture can allow users to ask questions about internal documents while grounding responses in organizational content. The PDF files can be processed and indexed, relevant passages can be retrieved for a user’s query, and those passages can be provided to a generative model as context. This architecture is useful when the source information is organization-specific and may change over time. Image classification and speech synthesis do not provide document-based question answering, while static HTML without retrieval does not provide the same dynamic retrieval and generation workflow.

Question 76

Which Azure AI Language capability can identify the overall emotional orientation of a text?

1. Sentiment analysis
2. OCR
3. Object detection
4. Image captioning

Correct Answer: 1

Explanation

Sentiment analysis determines the expressed sentiment of textual content. It can classify text into supported sentiment categories and provide confidence information that helps an application interpret the result. This can be useful for customer feedback, reviews, surveys, support messages, and other text-analysis scenarios. OCR extracts text from images, object detection identifies visual objects, and image captioning produces descriptions of images. Sentiment analysis should be treated as an automated analysis of textual language rather than a definitive assessment of a person’s actual feelings. Applications should evaluate results against their specific use case before relying on them operationally.

Question 77

Which component can be used to connect an AI application to external capabilities or data sources?

1. Tools
2. Image pixels
3. CSS styles
4. Screen resolution

Correct Answer: 1

Explanation

Tools can allow an AI application or agent to interact with external capabilities, services, or data sources. Depending on the architecture, a tool may perform actions such as retrieving information, querying a system, or executing an approved operation. This is especially relevant to agentic applications where the model needs more than the ability to generate text. Image pixels, CSS styles, and screen resolution do not provide general mechanisms for connecting an AI agent to external capabilities. Tool access should be carefully controlled using appropriate authentication, authorization, input validation, and monitoring to reduce security and operational risks.

Question 78

Why is grounding particularly useful for enterprise generative AI applications?

1. It provides access to organization-specific information
2. It guarantees every response is correct
3. It eliminates all security controls
4. It removes the need for evaluation

Correct Answer: 1

Explanation

Grounding allows a generative AI application to use organization-specific information that may not be represented in the model’s general knowledge. Enterprise applications often need to answer questions about internal policies, products, procedures, databases, or documents. Retrieval can provide relevant information from approved sources, which is then supplied to the model as context. Grounding does not guarantee that every generated response will be correct and does not eliminate the need for security, evaluation, or monitoring. It is instead a technique for improving the relevance and factual basis of responses using trusted external information.

Question 79

An application needs to convert a customer’s recorded phone call into text for later analysis. Which sequence should be used?

1. Speech-to-text, then language analysis
2. Image detection, then OCR
3. Text-to-speech, then translation
4. Object detection, then summarization

Correct Answer: 1

Explanation

Speech-to-text should first convert the recorded phone call into machine-readable text. Once the transcript is available, additional language-processing capabilities can analyze it for topics, sentiment, entities, summaries, or other required information. This creates a logical pipeline in which the audio-processing stage produces text for downstream AI services. Image detection and OCR are designed for visual content, while text-to-speech generates audio rather than transcribing it. Translation can be added if multilingual processing is required, but it is not the fundamental first step when the primary requirement is converting a recorded call into text.

Question 80

Which practice is important when deploying an AI solution that processes sensitive customer information?

1. Apply appropriate security and privacy controls
2. Publish customer data in prompts
3. Disable access restrictions
4. Store credentials in application text

Correct Answer: 1

Explanation

AI applications that process sensitive customer information should apply appropriate security and privacy controls throughout their architecture and lifecycle. These controls can include identity and access management, least-privilege permissions, secure handling of credentials, encryption, data protection, logging, and appropriate data-retention practices. Developers should also understand how data flows through connected AI services and ensure that processing aligns with organizational and regulatory requirements. Publishing customer data in prompts or storing credentials in application text can introduce significant risks. Security and privacy should therefore be treated as core design requirements rather than optional deployment steps.