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

 

Question 141

Which Azure AI capability can translate text from one language to another?

1. OCR
2. Object detection
3. Sentiment analysis
4. Machine translation

Correct Answer: 4

Explanation

Machine translation converts written content from one language into another supported language. It is useful for multilingual applications, customer-support systems, document processing, and communication platforms. Developers should consider language support, terminology, context, and quality requirements when implementing translation. Translation quality can vary depending on the language pair and content type, so testing with representative data is important. Machine translation is different from sentiment analysis, which evaluates the emotional tone of text, and OCR, which extracts text from visual documents.

Question 142

What is the primary purpose of Azure AI Search in a RAG solution?

1. Retrieve relevant information for the model
2. Generate speech
3. Detect objects in images
4. Create user passwords

Correct Answer: 1

Explanation

Azure AI Search can retrieve relevant information from indexed content and provide that information to a generative AI application. In a RAG architecture, documents are processed and indexed so user queries can identify useful content. The retrieved information can then be included as context for the language model. Azure AI Search supports different retrieval approaches, including keyword, semantic, vector, and hybrid search. Developers should configure indexing and retrieval carefully because the quality of retrieved context directly affects the quality of generated responses.

Question 143

Which feature allows an AI application to execute an external API through a defined operation?

1. OCR
2. Tool calling
3. Sentiment analysis
4. Text summarization

Correct Answer: 2

Explanation

Tool calling allows an AI application or agent to interact with external APIs and functions through defined operations. The model can determine that a tool is needed and provide the required arguments, while the application remains responsible for executing the operation. Tools can retrieve information, perform calculations, update records, or trigger business workflows. Developers should validate tool arguments and enforce authorization before execution. Sensitive operations may also require user confirmation or human approval to reduce the risk of unintended actions.

Question 144

Which approach is commonly used to improve retrieval when exact keywords are insufficient?

1. Audio processing
2. Semantic or vector search
3. Text-to-speech
4. Image compression

Correct Answer: 2

Explanation

Semantic or vector search can improve retrieval when exact keyword matching does not adequately capture the user’s intent. Vector search compares embeddings to identify content with similar meaning, while semantic search can use language understanding to improve result relevance. These approaches are useful when users phrase questions differently from the wording used in source documents. They can also be combined with keyword search through hybrid retrieval. Developers should evaluate retrieval quality using representative queries to determine which approach works best for their content.

Question 145

What does OCR primarily extract from an image or scanned document?

1. Spoken audio
2. Sentiment
3. Search permissions
4. Written or printed text

Correct Answer: 4

Explanation

Optical character recognition, or OCR, extracts written or printed text from images and scanned documents. The resulting text can then be searched, translated, summarized, classified, or passed to other information-extraction processes. OCR is commonly used for scanned forms, receipts, photographs of documents, and similar content. It does not analyze the emotional tone of text or convert spoken audio into words. Developers should consider image quality, resolution, handwriting support, and document layout when evaluating OCR performance for a specific application.

Question 146

Why should an AI developer define clear tool descriptions for an agent?

1. To help the model understand when and how to use each tool
2. To increase image resolution
3. To store passwords
4. To replace authorization checks

Correct Answer: 1

Explanation

Clear tool descriptions help an AI agent understand the purpose, inputs, outputs, and appropriate use of available tools. When an agent has several tools, accurate descriptions can improve its ability to select the correct operation for a task. Tool descriptions should explain required parameters and important constraints without being unnecessarily ambiguous. However, descriptions do not provide security by themselves. The application must still validate inputs, authenticate requests, authorize operations, and restrict tool permissions according to the intended workflow.

Question 147

Which capability can identify the main topics discussed in a document?

1. Key phrase extraction
2. Speech synthesis
3. Object detection
4. Image segmentation

Correct Answer: 1

Explanation

Key phrase extraction identifies important words and phrases that represent the main topics in a text. It can help developers quickly analyze large amounts of unstructured content such as reviews, reports, support tickets, and survey responses. The extracted phrases can be used for categorization, indexing, analytics, or content organization. This capability is provided through natural language processing services. It differs from named entity recognition, which focuses on identifying specific entities such as people, organizations, or locations.

Question 148

A developer wants to prevent users from accessing documents outside their department. What should be implemented?

1. Higher temperature
2. Access control and filtering
3. Larger embeddings
4. Text summarization

Correct Answer: 2

Explanation

Access control and appropriate filtering can restrict users to documents they are authorized to access. In an enterprise search or RAG system, document metadata can identify departments, security levels, or ownership. The application can use authenticated user information and authorization rules to filter retrieval results before they are provided to the model. This prevents unauthorized documents from becoming part of the model’s context. Access control should be enforced by application and service mechanisms rather than relying only on model instructions.

Question 149

Which capability can detect objects and their locations within an image?

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

Correct Answer: 1

Explanation

Object detection identifies objects within an image and provides information about their locations, commonly through bounding boxes or similar spatial representations. This differs from image classification, which generally assigns labels to an entire image without identifying the exact location of each object. Object detection can support applications such as inventory analysis, visual inspection, safety monitoring, and automated image processing. Developers should evaluate detection accuracy using images that represent the lighting, object sizes, backgrounds, and conditions expected in the production environment.

Question 150

What is a key advantage of using managed AI services?

1. Developers must build every model from scratch
2. They provide ready-to-use AI capabilities
3. They eliminate all application testing
4. They remove the need for authentication

Correct Answer: 2

Explanation

Managed AI services provide ready-to-use capabilities that developers can integrate into applications without building and operating every underlying AI model themselves. Examples include language processing, speech, vision, document analysis, and search services. These services can reduce development effort and provide scalable infrastructure managed by the cloud platform. Developers still need to configure the service, secure access, monitor usage, and evaluate results. Managed services do not eliminate the need for application design, testing, data protection, or responsible AI practices.

Question 151

What is the purpose of metadata associated with indexed documents?

1. Provide additional attributes for filtering and retrieval
2. Convert text into speech
3. Train every model automatically
4. Remove document permissions

Correct Answer: 1

Explanation

Metadata provides additional information about indexed documents that can be used to improve retrieval and filtering. Examples include document type, department, creation date, language, product category, or access classification. An application can use metadata to narrow search results or apply security-related filters. Metadata can also help organize and manage large collections of content. Developers should ensure that metadata is accurate and consistently maintained because incorrect metadata can lead to poor retrieval results or inappropriate document filtering.

Question 152

Which process converts spoken words into text?

1. Text-to-speech
2. Speech-to-text
3. OCR
4. Image analysis

Correct Answer: 2

Explanation

Speech-to-text converts spoken audio into written text. Azure AI Speech provides speech recognition capabilities that can be used for transcription, voice interfaces, meeting analysis, and other applications involving spoken language. The quality of transcription can be affected by background noise, speaker accents, audio quality, vocabulary, and language selection. Developers should test speech recognition using the types of audio expected in production. Speech-to-text differs from OCR, which extracts written characters from images or scanned documents.

Question 153

Why can retrieval quality affect the quality of a RAG response?

1. The model depends on retrieved context for grounded information
2. Retrieval controls microphone volume
3. Retrieval changes the user’s password
4. Retrieval creates computer vision models

Correct Answer: 1

Explanation

In a RAG system, retrieved information is commonly provided to the language model as context for generating an answer. If retrieval returns irrelevant, incomplete, outdated, or incorrect content, the model may not have the evidence needed to produce a useful response. Improving chunking, indexing, query processing, ranking, filtering, and embeddings can improve retrieval quality. Developers should evaluate retrieval independently as well as evaluating the final generated response. Good retrieval is therefore an important component of a reliable grounded AI application.

Question 154

Which practice helps an application recover from temporary service failures?

1. Ignoring all errors
2. Retry logic with appropriate delays
3. Removing monitoring
4. Disabling security controls

Correct Answer: 2

Explanation

Retry logic can help applications recover from temporary failures such as transient network problems, temporary service unavailability, or throttling. A controlled retry strategy typically includes an appropriate delay between attempts rather than immediately repeating requests continuously. Developers should also define a maximum number of retries and provide fallback behavior when retries are unsuccessful. Retry strategies should follow the relevant service guidance and consider whether repeating an operation could create duplicate side effects. Monitoring can help identify recurring failures and improve reliability.

Question 155

What is the main purpose of human approval in an AI workflow?

1. Allow a person to review an action before execution
2. Increase vector dimensions
3. Replace document indexing
4. Convert text into audio

Correct Answer: 1

Explanation

Human approval introduces a review step before an AI-generated action is executed. This can be useful when an agent is performing sensitive or consequential operations, such as sending communications, changing records, approving transactions, or making decisions that require human oversight. The application can present the proposed action and allow a person to approve, reject, or modify it. Human approval should complement technical controls such as authentication, authorization, input validation, and audit logging rather than being treated as the only security mechanism.

Question 156

Which factor should be considered when selecting a model for production?

1. Required quality, latency, cost, and capabilities
2. Screen size only
3. Keyboard type
4. File name length

Correct Answer: 1

Explanation

Model selection should consider the requirements of the application, including expected quality, supported capabilities, latency, cost, context size, throughput, and safety needs. A model that performs well for one task may not be appropriate for another. Developers should compare candidate models using representative inputs and measurable evaluation criteria. Production selection should also consider service availability, quotas, scaling requirements, and operational constraints. Choosing a model solely because it is larger or cheaper can result in an unsuitable solution.

Question 157

What can semantic search provide that exact keyword matching may miss?

1. Results based on related meaning
2. Automatic password rotation
3. Audio recording
4. Image resizing

Correct Answer: 1

Explanation

Semantic search can identify content that is conceptually related to a query even when the exact keywords are not present. For example, a user may ask about cancelling a subscription while a document uses different wording such as terminating a membership. Semantic retrieval can recognize the relationship between the concepts and return relevant information. This makes it useful for natural-language search and RAG applications. Developers should still evaluate semantic search with real queries because retrieval quality depends on the content and search configuration.

Question 158

Which approach can help reduce prompt injection risks in an agent that uses external tools?

1. Give every tool unrestricted permissions
2. Restrict tool permissions and validate inputs
3. Trust every retrieved instruction
4. Disable authentication

Correct Answer: 2

Explanation

Restricting tool permissions and validating inputs can reduce the potential impact of prompt injection attacks. An agent should receive only the permissions necessary for its intended tasks, and tool arguments should be checked before execution. Retrieved documents and user messages should be treated as potentially untrusted content rather than automatically trusted instructions. Sensitive operations can also require authorization or human approval. These controls provide stronger protection than relying on a system prompt alone and help prevent untrusted content from causing unauthorized actions.

Question 159

What does data residency generally refer to?

1. The geographic location where data is stored or processed
2. The number of model parameters
3. The speed of speech recognition
4. The size of an image

Correct Answer: 3

Explanation

Data residency generally refers to the geographic location in which data is stored or processed. Organizations may have requirements concerning where certain information can be stored because of legal, regulatory, contractual, or organizational policies. When designing an AI solution, developers should review the relevant service’s regional availability and data-handling documentation. Data residency is different from data retention, which concerns how long information is kept. Requirements can vary depending on the organization, type of information, and applicable regulations.

Question 160

Which practice helps ensure an AI application remains reliable after model updates?

1. Continuous evaluation and monitoring
2. Removing all test cases
3. Disabling logs
4. Allowing unrestricted model changes

Correct Answer: 1

Explanation

Continuous evaluation and monitoring can help identify changes in application behavior after a model or configuration is updated. A new model version may affect response quality, latency, cost, safety, or compatibility with existing prompts and tools. Developers can maintain representative evaluation datasets and compare important metrics before and after changes. Monitoring production behavior can identify issues that controlled testing did not reveal. Versioning, controlled deployment, and rollback procedures can further reduce the impact of unexpected changes.