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

 

Question 181

Which Azure AI Search capability retrieves documents using the semantic meaning of a query and indexed content?

1. Faceting
2. Vector search
3. Scoring profiles
4. Filtering

Correct Answer: 2

Explanation

Vector search retrieves content according to semantic similarity rather than depending only on matching individual words. An embedding model converts both the query and indexed content into numerical vectors that represent their meaning. Azure AI Search can then compare these vectors and retrieve content that is conceptually related to the query. This is useful when users phrase questions differently from the wording used in source documents. Vector search is commonly used in retrieval-augmented generation solutions and can be combined with keyword search, metadata filtering, and semantic ranking to improve the overall retrieval experience.

Question 182

A developer needs to identify the main topics discussed in customer feedback without creating predefined categories. Which capability should be used?

1. Custom classification
2. Language detection
3. Key phrase extraction
4. Entity recognition

Correct Answer: 3

Explanation

Key phrase extraction identifies important concepts and terms within unstructured text without requiring developers to define a fixed set of categories. It can help organizations analyze customer feedback, survey responses, reviews, support comments, and other text collections. For example, a customer message might produce phrases related to delivery delays, product quality, or customer service. This information can be used for analytics, indexing, or additional processing. Key phrase extraction differs from custom text classification because classification assigns predefined labels, whereas key phrase extraction identifies significant concepts present in the submitted text.

Question 183

Which capability allows an AI application to retrieve information from a private enterprise knowledge source before generating a response?

1. RAG
2. OCR
3. Speech synthesis
4. Image classification

Correct Answer: 1

Explanation

Retrieval-augmented generation, commonly called RAG, connects a generative AI model with an external knowledge source. The application first retrieves relevant information from sources such as enterprise documents, databases, or search indexes. That retrieved information is then supplied to the model as context for generating the response. RAG is useful when information changes frequently or contains organization-specific knowledge that is not part of the model’s original training data. A well-designed RAG system also needs appropriate access controls, effective chunking, reliable retrieval, grounding instructions, and evaluation procedures to maintain useful and trustworthy responses.

Question 184

Which Azure AI Speech capability converts written text into spoken audio?

1. Speech recognition
2. Speech translation
3. Text-to-speech
4. Pronunciation assessment

Correct Answer: 3

Explanation

Text-to-speech converts written text into synthesized spoken audio. Azure AI Speech can generate speech using supported voices and configuration options, making it useful for virtual assistants, accessibility applications, automated announcements, educational tools, and conversational systems. In a typical voice assistant, a language model may first generate a textual response, which is then passed to text-to-speech for audio output. Developers can select an appropriate voice and configure supported speech characteristics. Testing should consider pronunciation, language, speaking style, latency, and audio quality so that the synthesized response is suitable for the intended application.

Question 185

An organization wants to extract custom fields from its own contract documents. Which Azure AI Document Intelligence option is appropriate?

1. Prebuilt receipt model
2. Custom extraction model
3. Speech model
4. Sentiment model

Correct Answer: 2

Explanation

A custom extraction model in Azure AI Document Intelligence is appropriate when an organization needs to identify fields that are specific to its own document format. Contract documents may contain business-specific information such as agreement identifiers, renewal conditions, internal reference numbers, or specialized clauses. Developers can provide representative documents and label the required fields so that the model can learn the organization’s document structure. Prebuilt models are more appropriate when the document type and required fields are already supported. Custom extraction therefore provides flexibility when standardized document models cannot satisfy specialized business requirements.

Question 186

A developer wants search results to be restricted to documents belonging to the Finance department. Which Azure AI Search feature should be used?

1. Semantic captions
2. Filtering
3. Vector dimensions
4. Indexer scheduling

Correct Answer: 2

Explanation

Filtering allows Azure AI Search to restrict results according to values stored in structured fields. If each document contains a filterable Department field, an application can request only documents where the department equals Finance. Filters can also be used with other metadata such as document type, region, status, or access category. This makes filtering useful for narrowing search results before or alongside relevance-based retrieval. Developers should ensure that the required fields are configured as filterable during index design. Application-level authorization should still be enforced separately because a search filter alone should not be treated as a complete security boundary.

Question 187

Which Microsoft Foundry capability can help developers evaluate an AI application using predefined quality criteria and test data?

1. Model evaluation
2. OCR
3. Object detection
4. Speech synthesis

Correct Answer: 1

Explanation

Model evaluation helps developers assess an AI application’s behavior against defined criteria and representative test data. Depending on the scenario, evaluation can examine factors such as relevance, groundedness, response quality, safety, or other application-specific requirements. Structured evaluation is useful when comparing models, prompts, retrieval configurations, or application versions. It provides evidence that can guide improvements rather than relying only on subjective inspection of individual responses. Developers should create test datasets that reflect realistic user scenarios and include difficult cases. Evaluation should be repeated when significant changes are made to models, prompts, tools, or retrieval components.

Question 188

A generative AI agent must access a business application to retrieve customer order information. What should provide this capability?

1. A tool
2. A semantic caption
3. An OCR engine
4. A vector dimension

Correct Answer: 1

Explanation

An agent can use a tool to interact with an external business application such as an order-management system. The tool can expose a controlled operation that accepts parameters such as a customer or order identifier and returns the required information. This allows the agent to access current data instead of relying on information contained in the model’s pretrained knowledge. Tool permissions should be limited to necessary operations, and the external service should enforce authentication and authorization. Developers should also validate tool inputs and outputs and handle failures gracefully so that the agent does not provide misleading information when the business system is unavailable.

Question 189

Which Azure AI Language capability identifies entities such as organizations, locations, and people in text?

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

Correct Answer: 3

Explanation

Named entity recognition identifies recognized entity types within unstructured text. Common examples include people, organizations, locations, dates, quantities, and other supported categories. The extracted entities can be used to structure information that would otherwise remain embedded in natural-language documents. For example, a business application could identify organization names in customer messages and use them for routing or analysis. Standard named entity recognition is appropriate when the required entity types are supported by the service. Organizations with specialized terminology can consider custom named entity recognition when standard categories do not adequately represent their business requirements.

Question 190

A search solution needs both exact keyword matching and semantic similarity for the same query. Which approach should be selected?

1. OCR
2. Hybrid search
3. Speech translation
4. Classification

Correct Answer: 2

Explanation

Hybrid search combines traditional keyword retrieval with vector-based semantic retrieval. Keyword matching is useful for exact terms, identifiers, product names, and technical terminology, while vector retrieval can find content with similar meaning even when different words are used. Combining both approaches can provide stronger retrieval coverage for enterprise search and RAG applications. Azure AI Search can also apply additional ranking mechanisms to improve result ordering. Developers should evaluate hybrid retrieval using representative queries and source documents because the best configuration depends on factors such as content structure, query patterns, metadata, and the quality of the generated embeddings.

Question 191

Which Azure AI Vision capability can locate multiple objects within an image?

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

Correct Answer: 3

Explanation

Object detection identifies objects within an image and provides information about their locations. The output can include object labels along with bounding boxes indicating where the objects appear. This differs from image classification, which generally assigns labels to an image as a whole. Object detection can support applications such as retail inventory analysis, manufacturing inspection, traffic monitoring, and visual asset processing. Developers should select this capability when the application needs to know both what objects are present and where they are located. Accuracy should be evaluated using representative images from the intended production environment.

Question 192

A company wants its AI assistant to answer only from retrieved company documents and avoid unsupported information. Which instruction is useful?

1. Ignore all retrieved content
2. Use only supplied context
3. Always invent an answer
4. Disable retrieval

Correct Answer: 2

Explanation

An instruction to use only the supplied context can help guide a generative AI model toward grounded responses when the application provides retrieved company documents. The instruction can tell the model to rely on the retrieved evidence and clearly indicate when the available information does not support an answer. However, instructions alone do not guarantee reliable behavior. The application should also implement strong retrieval, appropriate access controls, evaluation, content safeguards, and validation where necessary. Grounding instructions are therefore one component of a larger RAG architecture designed to reduce unsupported or fabricated responses.

Question 193

Which Azure AI Search component defines enrichment operations such as entity recognition during document processing?

1. Skillset
2. Scoring profile
3. Query parser
4. Semantic configuration

Correct Answer: 1

Explanation

A skillset defines a collection of enrichment operations that can be applied during an Azure AI Search indexing pipeline. Skills can perform tasks such as text processing, OCR-related operations, entity recognition, language processing, and other supported enrichment activities. The resulting enriched information can then be mapped into fields in the search index. Skillsets are especially useful in knowledge-mining scenarios where raw documents need additional processing before they become useful for search or retrieval. Developers should design the enrichment pipeline according to the source data, required fields, processing costs, and downstream search requirements.

Question 194

A language-learning application needs to evaluate how accurately a user pronounces a provided sentence. Which Azure AI Speech capability should be used?

1. Speech translation
2. Pronunciation assessment
3. Text classification
4. Language detection

Correct Answer: 2

Explanation

Pronunciation assessment in Azure AI Speech is designed to evaluate spoken pronunciation against expected language content. It can support language-learning applications by providing feedback related to pronunciation and other supported speech characteristics. The application can provide reference text and capture the learner’s spoken response for analysis. This capability is different from speech translation, which focuses on converting spoken content into another language. Developers should test pronunciation assessment using representative speakers and recording environments because microphone quality, background noise, accents, and speech characteristics can affect recognition and assessment results.

Question 195

Which search technique can provide an exact nearest-neighbor comparison rather than using an approximate vector index?

1. Faceting
2. Semantic ranking
3. Exhaustive KNN
4. Keyword search

Correct Answer: 3

Explanation

Exhaustive KNN performs a direct nearest-neighbor comparison across the available vectors rather than relying on an approximate search structure. This can provide highly accurate similarity results because candidate vectors are directly compared with the query vector. The tradeoff is increased computational work, which can make exhaustive approaches less suitable for very large datasets or strict low-latency requirements. Approximate algorithms such as HNSW can provide faster retrieval at scale. Developers should select the approach based on collection size, accuracy expectations, performance requirements, and the characteristics of the application’s search workload.

Question 196

An application receives customer messages in unknown languages and must select an appropriate translation workflow. Which capability should be used first?

1. Language detection
2. Sentiment analysis
3. Object detection
4. Document extraction

Correct Answer: 1

Explanation

Language detection can determine which language is represented in incoming text before the application selects an appropriate translation or processing workflow. This is useful for multilingual customer-service systems where users may submit messages without identifying their language. After detection, the application can route the content to the required translation, classification, sentiment, or other language-processing service. Developers should account for ambiguous or very short messages because language identification may be less reliable when there is insufficient textual evidence. Supported languages and application requirements should also be considered when designing the overall multilingual pipeline.

Question 197

Which Azure AI Search feature can improve the ranking of keyword search results using language understanding?

1. Faceting
2. Semantic ranker
3. Indexer
4. Data source

Correct Answer: 2

Explanation

The semantic ranker can improve the ordering of candidate search results by applying language understanding to the query and retrieved documents. Traditional retrieval methods may identify relevant documents but do not always place the most useful result first. Semantic ranking can analyze the meaning and relationships within the candidate content to improve result ordering. It can be especially useful for natural-language questions and knowledge-search applications. Developers should configure the semantic search components appropriately and evaluate ranking quality with representative queries. Semantic ranking works as a ranking stage and does not replace the underlying search index or retrieval process.

Question 198

An application needs to extract invoice totals, vendor information, and line items from standard invoices. Which capability is suitable?

1. Custom text classification
2. Prebuilt invoice model
3. Speech-to-text
4. Image captioning

Correct Answer: 2

Explanation

The prebuilt invoice model in Azure AI Document Intelligence is designed to extract commonly required information from invoices. Depending on the document and supported model capabilities, this can include fields such as vendor information, invoice numbers, dates, totals, and line items. Using a prebuilt model can reduce development effort because the organization does not need to create and train an extraction model from scratch for a supported invoice scenario. If the organization’s invoices contain specialized fields that are not adequately handled by the prebuilt model, a custom extraction approach can be evaluated instead.

Question 199

A security team wants an AI application to detect potentially harmful prompts and outputs. Which capability should be incorporated into the solution?

1. Content safety
2. Vector indexing
3. Document classification
4. Speech recognition

Correct Answer: 1

Explanation

Content safety capabilities can help applications identify and manage potentially harmful content in user inputs and generated outputs. This is an important part of responsible generative AI development because models can encounter unsafe, abusive, or otherwise inappropriate requests and responses. Safety controls should be integrated into the application’s overall architecture rather than treated as an isolated feature. Developers should test expected use cases as well as adversarial or unusual inputs and establish appropriate handling policies. Access controls, monitoring, evaluation, and application-level safeguards should complement content safety mechanisms in production systems.

Question 200

A team deploys a new generative AI model and wants to determine whether its responses have become less accurate compared with the previous version. What should the team perform?

1. OCR processing
2. Regression evaluation
3. Image detection
4. Speech translation

Correct Answer: 2

Explanation

Regression evaluation compares a new model or application version against established test cases and expected quality criteria. By running the same representative prompts against the previous and new versions, developers can identify changes in response quality, groundedness, relevance, safety, or other important characteristics. This is particularly valuable for generative AI because model updates can alter behavior even when application code remains unchanged. A useful regression dataset should contain realistic production scenarios as well as challenging edge cases. Teams can use the results to identify unexpected degradation and investigate whether changes are needed before wider deployment.