View Full Microsoft AI-103 Exam Dumps and Practice Test Dumps.

 

Question 81

Which component is responsible for storing documents so they can be retrieved by an AI application?

1. Speech recognizer
2. Knowledge source
3. Text-to-speech engine
4. Image detector

Correct Answer: 2

Explanation

A knowledge source contains information that an AI application can use when answering questions or completing tasks. It may include documents, databases, websites, or other approved information repositories. In a retrieval-augmented generation solution, the application searches the knowledge source for relevant content and provides that information to the language model. This helps the model produce responses based on external information rather than relying only on its pretrained knowledge. Knowledge sources should be maintained and secured according to the application’s requirements.

Question 82

What does semantic search primarily use to determine document relevance?

1. Meaning and contextual similarity
2. File size
3. Creation date only
4. Document extension

Correct Answer: 1

Explanation

Semantic search evaluates the meaning and context of a query and compares it with available content. Unlike traditional keyword search, it can identify relevant information even when the exact words in the query do not appear in the document. Semantic search is especially useful for natural-language questions and knowledge retrieval applications. It can work with language understanding techniques and, in many AI architectures, complements vector search or keyword search. This helps users find conceptually relevant information instead of relying solely on exact term matching.

Question 83

An AI agent needs to check a customer’s order status. What should it use?

1. OCR
2. Image segmentation
3. A tool connected to the order system
4. Speech synthesis

Correct Answer: 3

Explanation

An AI agent can use a tool connected to an external order-management system to retrieve current order information. The tool can expose a controlled function or API that accepts required inputs, such as an order number, and returns the corresponding status. This approach allows the agent to access information that may change frequently without requiring the model to be retrained. Tool access should be authenticated and authorized, and the application should validate tool inputs and outputs before using the returned information.

Question 84

Which capability converts written text into spoken audio?

1. Speech-to-text
2. Text-to-speech
3. Optical character recognition
4. Named entity recognition

Correct Answer: 2

Explanation

Text-to-speech converts written text into synthesized spoken audio. Azure AI Speech provides text-to-speech capabilities that can be used in applications such as virtual assistants, accessibility solutions, announcements, and voice-based interfaces. Developers can select supported voices and configure speech characteristics depending on the application. Speech-to-text performs the reverse process by converting spoken audio into text. OCR extracts text from images or scanned documents, while named entity recognition identifies entities such as people, organizations, or locations within written language.

Question 85

Why might a developer choose a smaller language model for an application?

1. It always produces more accurate answers
2. It eliminates the need for testing
3. It removes all security risks
4. It may provide lower latency and cost

Correct Answer: 4

Explanation

Smaller language models can provide advantages such as lower latency, reduced resource requirements, and lower operating costs. They may be suitable for simpler tasks where a larger model is unnecessary. However, model selection should consider the application’s required quality, reasoning ability, context requirements, safety needs, and expected workload. A smaller model is not automatically better for every scenario. Developers should evaluate candidate models against representative test cases and production requirements before choosing the model for deployment.

Question 86

Which feature can help an AI application return data in a predictable schema?

1. Structured outputs
2. Object detection
3. Speech recognition
4. Image classification

Correct Answer: 1

Explanation

Structured outputs allow developers to request model responses that follow a defined structure or schema. This is useful when AI-generated information must be consumed programmatically by another application component. For example, an application could require fields such as customer name, order number, and status in a predictable JSON structure. The application should still validate the returned data before using it. Structured outputs improve consistency and reduce parsing problems compared with relying on unrestricted natural-language responses.

Question 87

What is the purpose of an AI model deployment?

1. To permanently train the model on every user request
2. To make a selected model available for application use
3. To convert documents into images
4. To create database backups

Correct Answer: 2

Explanation

A model deployment makes a selected AI model available through an endpoint or service configuration that an application can use. Deployment settings can include the chosen model, deployment name, capacity, and other service-specific options. Applications can then send requests to the deployed model rather than interacting with an undeployed model catalog entry. Developers should consider availability, performance, quotas, cost, authentication, and monitoring when deploying models for production workloads.

Question 88

An application receives an image and needs to identify multiple objects within it. Which capability is appropriate?

1. Sentiment analysis
2. Speech translation
3. Object detection
4. Text summarization

Correct Answer: 3

Explanation

Object detection identifies objects within an image and can provide information about where those objects appear. Depending on the service and model, the result may include object labels and bounding regions. This differs from image classification, which generally assigns one or more labels to an entire image without identifying the exact location of each object. Object detection is useful in scenarios such as inventory analysis, visual inspection, and image-based automation where the application needs to know both what objects are present and where they are located.

Question 89

What is a major benefit of using retrieval instead of retraining a model whenever company information changes?

1. It can use updated information without model retraining
2. It removes the need for authentication
3. It guarantees perfect responses
4. It eliminates all indexing requirements

Correct Answer: 1

Explanation

Retrieval allows an AI application to obtain current information from an external knowledge source at query time. When company documents change, the retrieval system can update its indexed content instead of requiring the language model itself to be retrained. This can make knowledge updates more practical for frequently changing information. The quality of the solution still depends on factors such as indexing, chunking, retrieval accuracy, grounding, and response generation. Retrieval also does not guarantee that every generated response will be correct.

Question 90

Which Azure AI capability can identify the emotional tone of written customer feedback?

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

Correct Answer: 4

Explanation

Sentiment analysis evaluates text to determine its expressed sentiment, such as positive, negative, or neutral, depending on the supported service and model. It can be useful for analyzing customer reviews, support messages, survey responses, and other written feedback. Azure AI Language provides natural language processing capabilities that support sentiment-related scenarios. Sentiment analysis is different from named entity recognition, which identifies entities, and key phrase extraction, which identifies important phrases. Developers should evaluate accuracy using representative domain-specific data.

Question 91

Which security approach is most appropriate for controlling access to AI resources?

1. Role-based access control
2. Publicly sharing credentials
3. Hard-coding administrator passwords
4. Removing authentication

Correct Answer: 1

Explanation

Role-based access control, or RBAC, allows organizations to assign permissions according to defined roles. Instead of giving every user or application unrestricted access, administrators can grant only the permissions required for specific tasks. RBAC can be used with Azure resources and services to control who can access, manage, or modify AI-related resources. Combining RBAC with managed identities and least-privilege principles can improve security. Permissions should be reviewed regularly to ensure that unnecessary access is removed.

Question 92

Why are content filters useful in generative AI applications?

1. They increase database storage
2. They help detect or restrict unsafe content
3. They create document embeddings
4. They translate source code

Correct Answer: 2

Explanation

Content filters can help identify or restrict certain categories of unsafe or inappropriate input and output in generative AI applications. They provide an additional safety layer around model interactions and can be configured according to the application’s requirements and supported service capabilities. Content filtering does not replace application security, authentication, authorization, or human oversight. Developers should evaluate safety controls using representative test cases and consider how the application should respond when content is blocked or flagged.

Question 93

A chatbot must answer questions from a large collection of company PDFs. What should be created first for efficient retrieval?

1. An audio transcription
2. A speech synthesis model
3. An indexed document collection
4. An image classification pipeline

Correct Answer: 3

Explanation

An indexed document collection allows the application to organize company content so relevant information can be retrieved efficiently. Before indexing, documents may need to be extracted, cleaned, chunked, enriched with metadata, and converted into searchable representations. Depending on the solution, the index can support keyword, semantic, vector, or hybrid search. During a user query, the application retrieves relevant sections and provides them to the language model. This architecture is commonly used for enterprise question-answering and RAG solutions.

Question 94

What should an AI application do before sending retrieved private documents to a language model?

1. Verify that access is authorized
2. Remove all authentication
3. Make the documents publicly available
4. Send every available document

Correct Answer: 1

Explanation

The application should verify that the user or service is authorized to access the retrieved information before providing private documents as model context. Retrieval systems must respect document-level permissions and other access controls. Simply finding a document through search does not automatically mean that every user should be allowed to view its contents. Security can include identity verification, authorization checks, metadata-based filtering, and role-based access control. These controls help prevent sensitive information from being exposed through AI-generated responses.

Question 95

Which task is most closely associated with Azure AI Language?

1. Object detection
2. Audio synthesis
3. Natural language processing
4. Image resizing

Correct Answer: 3

Explanation

Azure AI Language provides capabilities for processing and understanding human language. Depending on the service features and supported models, scenarios can include sentiment analysis, named entity recognition, key phrase extraction, summarization, language detection, and other text-based tasks. These capabilities can help developers add natural language processing to applications without building every language model from scratch. Azure AI Vision focuses primarily on visual content, while Azure AI Speech focuses on spoken language and audio-related scenarios.

Question 96

What is the purpose of monitoring an AI application after deployment?

1. To replace all testing
2. To track performance, errors, and usage
3. To remove application logging
4. To prevent every model response

Correct Answer: 2

Explanation

Monitoring helps developers understand how an AI application behaves after it is deployed. Useful signals can include request volume, latency, errors, token usage, failures, model performance indicators, and other application-specific metrics. Monitoring can reveal problems that were not visible during development or testing. Logs and telemetry can also help investigate incidents and identify opportunities for optimization. Monitoring should be designed carefully so that sensitive information is not unnecessarily captured in logs or exposed to unauthorized users.

Question 97

Which technique helps an AI system answer using the latest indexed company information?

1. Retrieval
2. Image compression
3. Audio normalization
4. Model temperature only

Correct Answer: 1

Explanation

Retrieval allows an AI system to obtain relevant information from an external source at the time a question is processed. If the company’s documents are updated and the search index is refreshed, future queries can retrieve the newer information. This is a key advantage of retrieval-augmented generation for enterprise knowledge scenarios. Developers should ensure that the indexing pipeline runs correctly and that stale documents are updated or removed. Retrieval quality also depends on appropriate search configuration, chunking, embeddings, and ranking.

Question 98

An AI agent must ask a user for approval before sending an email. Which design is appropriate?

1. Fully automatic execution without checks
2. Human-in-the-loop approval
3. Removing the email tool
4. Disabling authentication

Correct Answer: 2

Explanation

Human-in-the-loop approval introduces a human decision point before an important action is completed. For an AI agent that can send emails, the application can generate the proposed message and request user approval before invoking the email tool. This approach is useful for actions that may have business, financial, legal, or reputational consequences. The application should clearly show what action will occur and provide the user with an opportunity to approve, modify, or reject it before execution.

Question 99

What is the main purpose of prompt engineering?

1. Designing instructions that guide model responses
2. Replacing all application security
3. Creating physical network connections
4. Increasing document storage

Correct Answer: 1

Explanation

Prompt engineering involves designing and refining instructions and input structures to guide a generative AI model toward useful and consistent responses. Effective prompts can specify the task, context, desired format, constraints, and examples when appropriate. Developers can test different prompts using representative inputs and evaluate the resulting quality. Prompt engineering is useful but does not replace authentication, authorization, content filtering, or other security mechanisms. Applications should combine well-designed prompts with appropriate technical controls and validation.

Question 100

Which approach can reduce the impact of excessive requests sent to an AI service?

1. Removing request limits
2. Rate limiting
3. Sharing credentials
4. Disabling monitoring

Correct Answer: 2

Explanation

Rate limiting controls how many requests a user, application, or service can make during a defined period. It can help protect AI applications from excessive traffic, accidental request spikes, and certain forms of abuse. Rate limits can also help manage service quotas and operating costs. When a limit is reached, the application can return an appropriate response or use controlled retry behavior when suitable. Developers should choose limits based on expected traffic, service capabilities, performance requirements, and application needs.