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

 

Question 201

Which Azure AI Language capability can determine whether a sentence expresses positive or negative sentiment?

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

Correct Answer: 3

Explanation

Sentiment analysis evaluates text to determine its expressed emotional polarity, such as positive, negative, or neutral sentiment. It can be used for customer reviews, survey responses, support conversations, social feedback, and other textual data. The capability can provide useful insights into how users respond to products or services without requiring developers to manually inspect every message. Sentiment results should be evaluated against representative business content because sarcasm, ambiguous wording, mixed opinions, and domain-specific language can affect interpretation. Sentiment analysis is different from key phrase extraction, which identifies important concepts without determining emotional polarity.

Question 202

An application must find a customer’s name and company name in an incoming support message. Which capability should be considered?

1. Named entity recognition
2. OCR
3. Speech synthesis
4. Vector search

Correct Answer: 1

Explanation

Named entity recognition identifies recognized entity types within unstructured text. Depending on the service and supported categories, it can identify information such as people, organizations, locations, dates, and other entities. In a support application, the extracted customer and company names can be used for routing, personalization, indexing, or additional processing. Standard named entity recognition should be used when the required entity categories are supported. If the organization needs to identify specialized internal entities that are not covered by the standard model, custom named entity recognition can be trained with representative labeled examples.

Question 203

Which Azure AI Search capability is most suitable for finding documents that are conceptually similar to a user’s question?

1. Faceting
2. Filtering
3. Vector search
4. Indexer scheduling

Correct Answer: 3

Explanation

Vector search is designed to retrieve content according to semantic similarity. An embedding model converts the user’s question and indexed content into numerical vector representations. Azure AI Search then compares the query vector with stored vectors and retrieves content that is close in the configured vector space. This allows relevant documents to be discovered even when they use different wording from the user’s question. Vector search is commonly used in RAG applications and can be combined with keyword retrieval, filters, and semantic ranking. Developers should evaluate embedding quality, chunking strategy, and retrieval configuration using representative queries.

Question 204

A company wants to extract text and structured fields from scanned business documents. Which Azure service is designed for this task?

1. Azure AI Speech
2. Azure AI Document Intelligence
3. Azure AI Translator
4. Azure AI Language

Correct Answer: 2

Explanation

Azure AI Document Intelligence is designed to analyze documents and extract text, fields, tables, and other structured information. It supports several prebuilt document models and also provides custom model capabilities for specialized business documents. Scanned documents can be processed so that information that originally exists only visually becomes usable as structured data. This is useful for invoices, receipts, forms, applications, contracts, and similar workflows. Developers should choose the model according to the document type and extraction requirements. Representative documents should also be tested because layouts, scan quality, and field variations can affect extraction results.

Question 205

A developer needs an AI application to retrieve current information from an external service before answering a user. What should be configured?

1. A tool
2. An image classifier
3. A static prompt only
4. An OCR pipeline

Correct Answer: 1

Explanation

An external tool allows an AI application or agent to interact with a service that provides information not contained in the model’s static knowledge. For example, an agent could call a business API to obtain current inventory, account information, weather data, or another changing value. The tool should have clearly defined inputs and outputs and should be protected using appropriate authentication and authorization. Developers should also validate returned data before presenting it to users. Tool integration is particularly valuable for time-sensitive or transactional scenarios where relying only on a language model’s pretrained knowledge would not provide current information.

Question 206

Which Azure AI Search feature allows users to narrow results according to structured metadata?

1. Semantic captions
2. Filtering
3. Vector embeddings
4. Semantic ranking

Correct Answer: 2

Explanation

Filtering allows Azure AI Search queries to restrict results according to structured metadata fields. For example, an application can limit results to documents belonging to a particular department, document type, region, status, or access category. The relevant fields must be configured as filterable in the index schema. Filters can be combined with keyword, vector, or hybrid retrieval to narrow the candidate set according to application requirements. This is especially useful in enterprise search solutions where users need both relevance-based retrieval and metadata-based control. Developers should still enforce authorization separately when sensitive information is involved.

Question 207

A generative AI application must use internal documentation as evidence when answering user questions. Which architecture is appropriate?

1. Speech recognition
2. RAG
3. Image classification
4. OCR only

Correct Answer: 2

Explanation

Retrieval-augmented generation allows a generative AI application to retrieve relevant information from an external knowledge source and supply it to the model as context. For internal documentation, the documents can be processed, indexed, and made available through an appropriate retrieval system. When a user asks a question, relevant passages are retrieved and included in the model request. This can help the application provide answers based on company-specific information rather than relying only on pretrained knowledge. Developers should also address access permissions, retrieval quality, grounding instructions, document freshness, and evaluation when designing the complete RAG solution.

Question 208

Which Azure AI Speech capability converts spoken audio into text?

1. Text-to-speech
2. Speech-to-text
3. Pronunciation assessment
4. Speech synthesis markup

Correct Answer: 2

Explanation

Speech-to-text converts spoken audio into written text using speech recognition technology. It is useful for applications such as voice assistants, transcription systems, accessibility tools, meeting applications, and conversational interfaces. The resulting text can then be processed by language models, classification services, search systems, or other application components. Developers should select the appropriate source language and consider audio quality, background noise, microphone characteristics, and speaker behavior when evaluating recognition. Speech-to-text can also be paired with text-to-speech to create a complete two-way voice interaction where users speak to an application and receive spoken responses.

Question 209

A developer wants to prevent retrieved documents from unnecessarily consuming the model’s context window. What should be considered?

1. Context reduction
2. Image generation
3. Speech translation
4. Face analysis

Correct Answer: 1

Explanation

Context reduction can help limit the amount of retrieved information passed to a generative model. RAG systems may retrieve multiple documents or chunks, but not all retrieved content is equally useful for answering the user’s question. Reducing irrelevant context can lower token consumption, decrease latency, and make it easier for the model to focus on important evidence. Techniques can include stronger ranking, metadata filtering, smaller chunks, relevance thresholds, and summarization or compression. Developers should measure whether context reduction maintains answer quality because removing useful evidence can negatively affect grounding and completeness.

Question 210

Which Microsoft Foundry capability can help developers trace operations across a multi-step AI workflow?

1. Tracing
2. Faceting
3. OCR
4. Image tagging

Correct Answer: 1

Explanation

Tracing provides visibility into the sequence of operations performed during an AI application workflow. In a multi-step generative AI application, developers may need to understand when prompts were sent, which model was called, what retrieval operation occurred, or which tool was invoked. Trace information can help identify failures, unexpected behavior, latency problems, and inefficient workflows. It is particularly valuable for applications containing agents and multiple connected components. Tracing should be combined with evaluation and monitoring to understand both individual execution paths and broader application performance over time.

Question 211

Which capability is designed to detect potentially harmful content in generative AI inputs and outputs?

1. Vector search
2. Content safety
3. Document extraction
4. Language detection

Correct Answer: 2

Explanation

Content safety capabilities help identify and manage potentially harmful content in AI application inputs and outputs. This is important for generative AI systems because users may submit unsafe requests and models may generate content that violates application policies. Safety controls can be integrated into application workflows to assess content and determine whether it should be allowed, blocked, transformed, or reviewed. Developers should test safety mechanisms with representative and adversarial examples because no single control handles every possible scenario. Content safety should also be combined with authentication, authorization, monitoring, responsible AI practices, and application-specific policies.

Question 212

A developer needs to classify incoming documents before selecting the appropriate extraction model. Which capability should be considered?

1. Document classification
2. Speech synthesis
3. Sentiment analysis
4. Vector embedding

Correct Answer: 1

Explanation

Document classification can determine which category a document belongs to before it enters a specialized processing workflow. For example, an intake system might classify incoming files as invoices, receipts, contracts, or application forms. The classification result can then determine which Document Intelligence model or downstream workflow should process the file. This approach is useful when multiple document types arrive through a shared channel. Developers should provide representative examples for the expected categories and evaluate classification accuracy. Clear category definitions can also reduce ambiguity when documents have similar layouts or overlapping content.

Question 213

Which Azure AI Search feature combines lexical retrieval and vector retrieval in a single search experience?

1. Hybrid search
2. OCR
3. Faceting
4. Scoring profile

Correct Answer: 1

Explanation

Hybrid search combines keyword-based lexical retrieval with vector-based semantic retrieval. This allows an application to benefit from exact term matching while also retrieving content that is conceptually similar to the query. For example, a technical support search may need to recognize an exact product code while also finding documents that describe the same problem using different language. Hybrid retrieval can therefore improve coverage across different query types. Developers can further apply semantic ranking, filters, and other search features depending on the application. Testing should use representative queries to determine whether the combined approach improves retrieval quality.

Question 214

An application needs to identify the exact location of detected objects in an image. Which capability is appropriate?

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

Correct Answer: 2

Explanation

Object detection identifies objects within an image and provides their locations, commonly through bounding boxes. This makes it different from image classification, which generally assigns categories to an image without identifying individual object positions. Object detection can support manufacturing inspection, retail inventory analysis, safety monitoring, visual search, and other computer vision scenarios. The application can use the detected object labels and coordinates for subsequent processing or business logic. Developers should evaluate the model using images that reflect actual production conditions, including different lighting, object sizes, backgrounds, orientations, and levels of visual complexity.

Question 215

Which Azure AI Language capability can extract important concepts from a customer review without assigning a predefined category?

1. Custom text classification
2. Sentiment analysis
3. Key phrase extraction
4. Custom entity recognition

Correct Answer: 3

Explanation

Key phrase extraction identifies significant concepts and terms within text without requiring predefined classification labels. It can be useful for analyzing customer reviews, survey responses, support messages, or large collections of unstructured content. For example, an application could identify phrases related to delivery, product quality, customer service, or pricing and use them for further analysis. This differs from custom text classification, where developers define categories and train the model to assign those categories. Key phrase extraction can also complement sentiment analysis by showing which topics are being discussed while sentiment analysis evaluates the emotional polarity associated with the text.

Question 216

A developer needs to store numerical embeddings in an Azure AI Search index for similarity retrieval. Which field should be configured?

1. Searchable text field only
2. Vector field
3. Date field
4. Boolean field

Correct Answer: 2

Explanation

A vector field is used to store numerical embeddings that support vector search in Azure AI Search. The embeddings can represent documents, document chunks, queries, images, or other supported content depending on the application. The vector field must be configured according to the embedding dimensions and the selected vector search configuration. Other fields can store the original text and metadata needed for filtering or displaying results. Keeping source content and metadata alongside vectors is useful because retrieved vectors themselves are primarily used for similarity calculations, while the associated fields provide meaningful information for the final application response.

Question 217

Which Azure AI Speech capability can control pronunciation, pauses, and emphasis in synthesized speech?

1. SSML
2. OCR
3. Vector search
4. Entity recognition

Correct Answer: 1

Explanation

Speech Synthesis Markup Language, or SSML, provides structured controls for speech synthesis. Developers can use supported SSML elements to influence characteristics such as pronunciation, pauses, emphasis, speaking rate, pitch, and other aspects of synthesized speech. This is useful when the default speech output does not provide the desired pronunciation or presentation. For example, an application may need a product name pronounced in a particular way or may need a pause between important statements. Developers should test SSML with the selected voice and language because supported features and behavior can vary between speech configurations.

Question 218

A company has domain-specific entity types that standard named entity recognition cannot identify reliably. What should the company consider?

1. Custom named entity recognition
2. Sentiment analysis
3. Language detection
4. Speech translation

Correct Answer: 1

Explanation

Custom named entity recognition allows organizations to train a model to identify domain-specific entities that are not adequately covered by standard entity recognition. Examples can include proprietary product names, internal project codes, specialized medical terminology, or industry-specific identifiers. Developers provide labeled training data showing how these custom entities appear in real text. The trained model can then process new content and identify the defined entity types. High-quality training examples are important because the model needs to learn the terminology and variations used in the organization’s actual documents. Evaluation should use separate representative data rather than only training examples.

Question 219

A team wants to compare two prompts using the same set of representative user questions. Which approach should be used?

1. Prompt evaluation
2. Image segmentation
3. OCR extraction
4. Speech translation

Correct Answer: 1

Explanation

Prompt evaluation allows developers to assess how different prompt designs perform against the same set of representative inputs. Keeping the test questions consistent makes it easier to identify whether a change in instructions affects response quality, relevance, groundedness, or other selected criteria. Prompt evaluation is particularly useful during generative AI development because small instruction changes can influence model behavior. Developers should include common scenarios, edge cases, and potentially difficult inputs in the evaluation dataset. Results should be interpreted using clearly defined criteria rather than relying only on a small number of manually inspected responses.

Question 220

An AI agent needs permission to read a database but should not be allowed to modify records. Which security principle should guide the permission design?

1. Full access
2. Anonymous access
3. Least privilege
4. Shared credentials

Correct Answer: 3

Explanation

Least privilege means granting an identity only the permissions necessary to perform its intended tasks. If an AI agent only needs to read database records, it should not receive permissions that allow inserts, updates, or deletions. Restricting permissions reduces the potential impact of incorrect tool calls, compromised credentials, prompt manipulation, or application errors. Access should be implemented through appropriate authentication and authorization mechanisms, with permissions reviewed periodically. For agent-based systems, this principle is especially important because tools can allow models to interact with real systems. The agent’s available operations should therefore be limited to approved business requirements.