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

 

Question 101

Which Azure AI Language capability can identify the language used in a text input?

1. Entity linking
2. Language detection
3. Sentiment analysis
4. Summarization

Correct Answer: 2

Explanation

Language detection determines which supported language is represented by a text input. This capability is useful when an application receives content from users who may communicate in different languages and the application needs to select an appropriate processing workflow. For example, detected language information can be used to route content for translation or language-specific analysis. Entity linking enriches recognized entities, sentiment analysis evaluates expressed sentiment, and summarization creates shorter versions of content. Language detection should therefore be selected when the application’s primary requirement is identifying the language of submitted textual content before additional processing occurs.

Question 102

A developer needs to extract text from a photograph of a menu. Which capability should be used?

1. Sentiment analysis
2. Semantic ranking
3. OCR
4. Entity recognition

Correct Answer: 3

Explanation

Optical character recognition, or OCR, extracts machine-readable text from images. A photograph of a menu can contain printed words that are not initially available as digital text. OCR can process the image and return recognized text, which can then be translated, searched, summarized, or analyzed by other services. Sentiment analysis and entity recognition operate on textual information after it is available, while semantic ranking is related to search result relevance. Therefore, OCR is the appropriate first capability when the application needs to convert visible menu text into information that software can process.

Question 103

Which capability can help an application determine whether a review is positive, negative, or neutral?

1. Object detection
2. Speech recognition
3. Sentiment analysis
4. OCR

Correct Answer: 3

Explanation

Sentiment analysis evaluates textual content and can classify its expressed sentiment into supported categories such as positive, negative, or neutral. It can also provide confidence information that helps an application interpret the result. This capability is useful for analyzing product reviews, surveys, customer comments, and support interactions. Object detection processes visual objects, speech recognition converts audio into text, and OCR extracts text from images. Sentiment analysis does not guarantee an exact understanding of a person’s feelings, so results should be evaluated in the context of the application. Nevertheless, it is the appropriate capability for automated review sentiment analysis.

Question 104

What is the primary role of an indexer in Azure AI Search?

1. Define user permissions
2. Generate model responses
3. Store application secrets
4. Import and update searchable content

Correct Answer: 4

Explanation

An Azure AI Search indexer automates the process of retrieving content from a supported data source and adding or updating that content in a search index. This can reduce the amount of custom ingestion code required for supported data sources. The indexer can also work with a configured skillset when content enrichment is needed during indexing. An index stores searchable representations, while the indexer performs the ingestion process. User permissions, model response generation, and secret management are separate concerns. Understanding the indexer’s role is important when building maintainable search and knowledge-mining pipelines.

Question 105

A generative AI application should answer using information retrieved from an approved knowledge base. What should the developer configure?

1. Grounding instructions and retrieved context
2. Image resizing
3. Audio compression
4. DNS routing

Correct Answer: 1

Explanation

Grounding instructions and retrieved context can help a generative AI application produce responses based on approved knowledge sources. The application can retrieve relevant passages from the knowledge base and provide them to the model with instructions explaining how the information should be used. It can also specify what the model should do when the context does not contain enough information. This approach can improve factual relevance but does not guarantee perfect answers. Image resizing, audio compression, and DNS routing address unrelated technical requirements and do not provide the model with trusted knowledge-base content.

Question 106

Which Azure AI Search capability is designed to retrieve documents using semantic similarity?

1. Vector search
2. Speech recognition
3. OCR
4. Text-to-speech

Correct Answer: 1

Explanation

Vector search retrieves content by comparing vector representations of queries and indexed documents. Because these vectors can capture semantic characteristics, vector search can identify conceptually related information even when the exact keywords differ. This is particularly useful for knowledge bases and retrieval-augmented generation applications. Speech recognition converts audio to text, OCR extracts text from images, and text-to-speech converts written text into audio. A vector search implementation depends on suitable embeddings and index configuration. Developers should evaluate retrieval quality using representative queries to ensure that semantically relevant content is being returned.

Question 107

Which model type is most appropriate for generating an image from a textual description?

1. Speech model
2. Vision classifier
3. Image generation model
4. Search index

Correct Answer: 3

Explanation

An image generation model creates visual content based on a textual or other supported prompt. In Microsoft’s AI ecosystem, supported image-generation models can be used for scenarios such as creating illustrations, concepts, or other generated visual assets. A speech model handles audio and spoken language, a vision classifier categorizes existing images, and a search index retrieves stored information. Image generation should be evaluated for quality, safety, content requirements, and application-specific constraints. The important distinction is that generation creates new visual content, whereas classification and analysis operate on content that already exists.

Question 108

A developer wants to compare two prompt versions to determine which produces better results. What should be used?

1. Network monitoring
2. Model evaluation
3. File compression
4. Database replication

Correct Answer: 2

Explanation

Model evaluation can be used to compare the outputs produced by different prompts or configurations against predefined criteria. Developers can run representative test inputs through each prompt version and assess factors such as relevance, correctness, groundedness, consistency, and safety. This provides evidence for deciding which configuration better satisfies the application’s requirements. Network monitoring, file compression, and database replication do not directly measure prompt or model quality. Evaluation should use a representative dataset rather than a single example, because generative models can behave differently across different inputs and edge cases.

Question 109

Which Azure AI Language capability can identify important concepts within a document without producing a full summary?

1. Speech translation
2. Key phrase extraction
3. Object detection
4. Image captioning

Correct Answer: 2

Explanation

Key phrase extraction identifies important words and phrases that represent major concepts in a text. It can provide a compact representation of document topics without generating a complete natural-language summary. For example, an application processing customer feedback might extract phrases related to delivery, pricing, product quality, and technical support. Speech translation works with spoken language, object detection identifies objects in images, and image captioning describes visual content. Key phrase extraction is therefore appropriate when the requirement is to identify significant concepts for indexing, categorization, analysis, or search rather than produce a full summary.

Question 110

Which security principle limits an application’s permissions to only what it needs?

1. Semantic ranking
2. Prompt chaining
3. Data augmentation
4. Least privilege

Correct Answer: 4

Explanation

Least privilege means granting an identity, application, or component only the permissions required to perform its intended tasks. This principle reduces unnecessary access and limits the potential impact of a compromised identity or application. For AI solutions, least privilege can be applied to access storage, databases, search indexes, model resources, and other connected services. Semantic ranking, prompt chaining, and data augmentation are AI or search concepts rather than access-control principles. Implementing least privilege should be combined with authentication, authorization, secure credential management, logging, and regular access reviews to maintain an appropriate security posture.

Question 111

Which feature can improve search results by understanding the meaning of a user’s query?

1. Semantic ranking
2. OCR
3. Speech synthesis
4. Image segmentation

Correct Answer: 1

Explanation

Semantic ranking uses language understanding to improve the ordering of search results according to the meaning of the user’s query and the content of candidate documents. It can be applied after initial retrieval to help place more relevant results higher in the result set. OCR extracts text from images, speech synthesis produces spoken audio, and image segmentation identifies regions within visual content. Semantic ranking is particularly useful when keyword matching alone does not adequately distinguish the most relevant documents. Search quality should still be tested using representative queries and expected results.

Question 112

An application receives voice commands from users. Which capability converts those commands into text for processing?

1. Text-to-speech
2. Speech-to-text
3. Image captioning
4. Document extraction

Correct Answer: 2

Explanation

Speech-to-text converts spoken audio into written text that an application can process. This capability is commonly used in voice assistants, customer-service systems, meeting transcription, accessibility applications, and other voice-driven solutions. Once the speech has been converted into text, the application can send the text to a language model or other language-processing service. Text-to-speech performs the opposite direction by generating spoken audio from text. Image captioning and document extraction address visual and document-processing scenarios. Therefore, speech-to-text is the appropriate capability for converting voice commands into machine-readable text.

Question 113

Which approach is useful when a model must access current information that changes frequently?

1. Static model knowledge
2. Retrieved external data
3. Deleted search indexes
4. Fixed responses

Correct Answer: 2

Explanation

Retrieved external data can provide a model with current information that changes more frequently than the model’s underlying knowledge. For example, product availability, internal policies, current schedules, or inventory data can be retrieved from an authoritative source at query time. The retrieved information can then be supplied to the model as grounding context. Relying exclusively on static model knowledge may result in outdated responses. Deleted indexes and fixed responses cannot provide current information. A retrieval-based architecture should also include access controls and validation so that the model receives information from appropriate and trusted sources.

Question 114

Which component defines the fields and searchable structure of an Azure AI Search index?

1. Search index
2. Speech model
3. Language detector
4. Image classifier

Correct Answer: 1

Explanation

An Azure AI Search index defines the searchable structure of indexed content, including fields and their characteristics. Depending on the solution, fields can support searching, filtering, sorting, faceting, semantic capabilities, or vector retrieval. An indexer can populate the index from a supported data source, while a skillset can enrich content during the indexing pipeline. Speech models, language detectors, and image classifiers serve different AI workloads. The index is therefore the component that establishes how searchable content is represented and queried. Careful schema design is important because it affects retrieval behavior and application capabilities.

Question 115

A developer wants an agent to retrieve information from an external API. What should be provided to the agent?

1. A tool with an appropriate interface
2. A larger image
3. A CSS stylesheet
4. A speech codec

Correct Answer: 1

Explanation

An agent can use a tool to interact with an external API through a defined interface. The tool describes the operation available to the agent and provides the mechanism for sending appropriate inputs and receiving results. For example, an order lookup tool could accept a customer or order identifier and return information from a business system. External tool access should be secured and limited to authorized operations. Larger images, CSS stylesheets, and speech codecs do not provide an interface for calling external business APIs. Tool integration therefore extends an agent’s capabilities beyond language generation.

Question 116

Which document-processing capability can identify fields from invoices?

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

Correct Answer: 3

Explanation

Azure AI Document Intelligence provides document-analysis capabilities that can extract text, fields, tables, and other structured information from supported documents. Invoice processing is a common business scenario where fields such as vendor information, invoice identifiers, dates, totals, and line items may need to be extracted automatically. Image captioning describes images, sentiment analysis evaluates textual sentiment, and speech translation processes spoken language. Document Intelligence can use prebuilt models for supported document scenarios and custom models for specialized requirements. Extracted results should be validated against representative invoices before being used in important automated business processes.

Question 117

What is a major benefit of hybrid search in Azure AI Search?

1. It uses only exact keywords
2. It combines lexical and semantic retrieval
3. It eliminates the need for indexes
4. It converts documents into audio

Correct Answer: 2

Explanation

Hybrid search combines keyword-based retrieval with vector-based semantic retrieval. Lexical search can be effective when exact terms are important, while vector retrieval can find conceptually related information even when different wording is used. Combining both approaches can improve retrieval quality for applications such as enterprise search and retrieval-augmented generation. Hybrid search does not eliminate indexes or convert documents into audio. It also does not guarantee perfect relevance, so developers should evaluate the results using representative queries and tune search configuration according to the application’s information-retrieval requirements.

Question 118

Which action is appropriate when an AI application produces unsafe or inappropriate output during testing?

1. Ignore the output
2. Remove evaluation
3. Investigate and adjust safety controls
4. Publish the application immediately

Correct Answer: 3

Explanation

Unsafe or inappropriate output discovered during testing should be investigated before production deployment. Developers can examine the prompt, model configuration, input data, content filtering, grounding, and application logic to determine why the undesirable response occurred. Appropriate safety controls and instructions can then be adjusted and the scenario retested. Ignoring the issue or publishing the application immediately can expose users and the organization to unnecessary risk. Removing evaluation would also reduce visibility into the problem. Safety testing should include representative and adversarial scenarios so that important failure modes can be identified and addressed.

Question 119

A developer needs to determine the sentiment of text that has first been extracted from a scanned document. Which sequence is appropriate?

1. Text-to-speech then OCR
2. OCR then sentiment analysis
3. Speech translation then object detection
4. Image classification then speech synthesis

Correct Answer: 2

Explanation

OCR should first extract the textual content from the scanned document. Once the text is available in machine-readable form, Azure AI Language sentiment analysis can evaluate its expressed sentiment. This two-stage process is appropriate because sentiment analysis operates on text rather than the original scanned image. Text-to-speech generates audio, speech translation processes spoken language, and object detection identifies visual objects. Image classification can categorize images but does not directly determine textual sentiment. Combining specialized AI capabilities in a pipeline allows an application to transform visual source material into analyzed textual information.

Question 120

Why should AI applications monitor model and service behavior after deployment?

1. To identify operational or quality changes
2. To remove authentication
3. To stop collecting telemetry
4. To avoid all testing

Correct Answer: 1

Explanation

Monitoring after deployment helps identify changes in model behavior, service performance, resource usage, errors, latency, and other operational characteristics. AI applications can encounter real-world inputs and workloads that were not fully represented during development. Monitoring provides visibility that can help teams detect unexpected behavior and investigate potential problems. It does not replace predeployment testing, and authentication and security controls should remain enabled. Effective monitoring can also support continuous improvement by providing evidence about how the application performs in production. Therefore, ongoing observation is an important part of operating a reliable AI solution.