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

 

Question 101

Which Azure AI capability can extract key phrases from customer reviews?

1. Azure AI Language
2. Azure AI Vision
3. Azure AI Speech
4. Azure AI Search

Correct Answer: 1

Explanation

Azure AI Language provides natural language processing capabilities that can identify important words and phrases within text. Key phrase extraction can help summarize the main topics discussed in customer reviews, survey responses, support tickets, and other documents. For example, a review about delivery delays and damaged packaging could produce phrases related to those topics. Developers can use these extracted phrases to organize content, identify recurring issues, or support downstream analytics. This capability works with text rather than images or audio.

Question 102

What is the main purpose of an indexer in Azure AI Search?

1. Generate speech from text
2. Extract and load content into a search index
3. Create language model prompts
4. Detect objects in photographs

Correct Answer: 2

Explanation

An indexer in Azure AI Search automates the process of extracting information from supported data sources and adding it to a search index. It can connect to sources such as storage services and process available content according to the configured indexing pipeline. This reduces the need to write custom code for repeatedly loading documents into an index. Once content is indexed, applications can use keyword, semantic, vector, or hybrid search to retrieve relevant information for users or AI applications.

Question 103

Which technique represents documents as vectors for similarity-based retrieval?

1. OCR
2. Speech recognition
3. Embedding generation
4. Text-to-speech

Correct Answer: 3

Explanation

Embedding generation converts content such as text into numerical vectors that represent semantic characteristics of the content. These vectors can then be stored in a vector-enabled search system and compared with vectors generated from user queries. Content with similar meanings can have similar vector representations even when the exact words differ. This makes embeddings useful for semantic retrieval, recommendation systems, and RAG applications. The quality of retrieval depends on factors such as the embedding model, chunking strategy, index configuration, and query processing.

Question 104

An application must summarize long documents before presenting them to users. Which AI capability is most relevant?

1. Object detection
2. Speech synthesis
3. Named entity recognition
4. Text summarization

Correct Answer: 4

Explanation

Text summarization generates a shorter representation of longer text while attempting to preserve important information. It can be useful for summarizing reports, customer conversations, articles, support tickets, or business documents. Depending on the scenario, developers may use extractive or generative approaches. The resulting summary should still be evaluated for accuracy because a model can omit important details or produce unsupported statements. For high-impact applications, developers can combine summarization with grounding, validation, and human review when appropriate.

Question 105

Why is chunk overlap sometimes used when splitting documents?

1. To preserve context between adjacent chunks
2. To encrypt the document
3. To increase image resolution
4. To convert speech into text

Correct Answer: 1

Explanation

Chunk overlap means that a portion of one document chunk is repeated in the following chunk. This can help preserve context when an important sentence, explanation, or concept crosses a chunk boundary. Without overlap, related information might be separated in a way that makes individual chunks less useful during retrieval. The amount of overlap should be selected carefully because excessive overlap can increase the number of tokens, storage requirements, and duplicate retrieval results. Chunk size and overlap should be tested against representative documents.

Question 106

Which feature allows an AI application to process information from images and text together?

1. Batch processing
2. Multimodal AI
3. Rate limiting
4. Semantic indexing

Correct Answer: 2

Explanation

Multimodal AI can process or generate information involving multiple types of input, such as text and images. For example, a multimodal application could receive a photograph of a product along with a written question and use both sources when generating a response. This capability is useful for visual question answering, document understanding, product analysis, and other scenarios that require more than text alone. Developers should verify that the selected model supports the required modalities and evaluate its performance with realistic multimodal inputs.

Question 107

What should an application use to securely store API keys and other secrets?

1. Azure Key Vault
2. Public source code
3. A plain-text document
4. Client-side HTML

Correct Answer: 1

Explanation

Azure Key Vault is designed to securely store and manage secrets, keys, and certificates. Using a dedicated secret-management service helps prevent sensitive credentials from being embedded directly in source code or configuration files. Applications can use managed identities and appropriate access permissions to retrieve secrets when necessary. Developers should also rotate credentials when required and limit access according to the principle of least privilege. Keeping secrets out of source repositories reduces the risk of accidental exposure through code sharing, logs, or version-control systems.

Question 108

Which search method combines keyword matching with vector similarity?

1. Semantic-only search
2. Hybrid search
3. OCR search
4. Speech search

Correct Answer: 2

Explanation

Hybrid search combines traditional keyword-based retrieval with vector-based similarity search. Keyword matching is useful for exact terms, product names, identifiers, and other precise text, while vector search can identify content with similar meaning even when different wording is used. Combining the two can provide broader retrieval coverage for enterprise AI applications. Azure AI Search supports hybrid search scenarios and can be used as part of RAG architectures. Developers should evaluate the retrieval configuration using realistic queries and representative documents.

Question 109

A model repeatedly produces answers that are not supported by retrieved documents. What should the developer investigate first?

1. Grounding and retrieval quality
2. Screen resolution
3. Audio volume
4. Image compression

Correct Answer: 1

Explanation

When generated answers are not supported by retrieved information, developers should investigate the grounding and retrieval pipeline. Important factors include whether the correct documents are indexed, whether chunking is appropriate, whether search returns relevant content, and whether the retrieved context is actually provided to the model. Developers should also check prompts and response validation. Improving retrieval quality can give the model better evidence for generating an answer. Evaluation with representative questions can help determine whether changes to the retrieval pipeline improve groundedness.

Question 110

Which capability identifies the language used in a text input?

1. Object detection
2. Language detection
3. Speech synthesis
4. Image classification

Correct Answer: 2

Explanation

Language detection identifies the language represented in a text input. This can be useful in applications that receive content from users in multiple languages and need to route the text to the appropriate processing workflow. For example, an application could detect whether a message is written in English, Urdu, Arabic, or another supported language before performing translation or language-specific analysis. Language detection is a natural language processing capability and differs from speech recognition, which processes spoken audio.

Question 111

What is the purpose of query rewriting in a search-based AI application?

1. Improve a user’s query before retrieval
2. Encrypt the search index
3. Convert images into audio
4. Create user accounts

Correct Answer: 1

Explanation

Query rewriting modifies or expands a user’s original query to improve retrieval results. A user may ask a vague question, use conversational language, or omit important search terms. A query-rewriting process can transform the request into a form that better matches the indexed content. This can be especially useful in RAG applications where retrieval quality directly affects the context provided to the language model. Developers should evaluate rewritten queries carefully to ensure that the intended meaning of the user’s request is preserved.

Question 112

Which capability is used to recognize text printed on a scanned document?

1. Text-to-speech
2. Optical character recognition
3. Sentiment analysis
4. Speech translation

Correct Answer: 2

Explanation

Optical character recognition, or OCR, identifies printed or handwritten characters in supported images and documents and converts them into machine-readable text. OCR is commonly used when applications need to process scanned forms, photographed documents, receipts, or other visual content containing text. The extracted text can then be searched, analyzed, translated, or passed to additional information-extraction processes. OCR is different from speech-to-text because OCR processes visual content, while speech-to-text converts spoken audio into written language.

Question 113

Which approach is appropriate when an AI agent must perform an action in an external business system?

1. Tool calling
2. Image classification
3. OCR only
4. Sentiment analysis

Correct Answer: 1

Explanation

Tool calling allows an AI agent to interact with external systems through controlled functions or APIs. For example, an agent could use a tool to create a support ticket, retrieve an order, check inventory, or update a business record. The application should define clear tool inputs and outputs and enforce authentication and authorization before execution. For sensitive actions, human approval may also be appropriate. Tool calling extends an agent’s capabilities beyond generating text and enables it to interact with real-world business workflows.

Question 114

What is the primary purpose of model evaluation?

1. Increase storage capacity
2. Measure how well a model performs against defined requirements
3. Replace access control
4. Remove all model limitations

Correct Answer: 2

Explanation

Model evaluation measures how well an AI model or application performs against defined requirements and representative test cases. Depending on the scenario, evaluation can examine accuracy, relevance, groundedness, safety, latency, or other quality metrics. Developers can compare results across models, prompts, or configurations to identify weaknesses. Evaluation should be performed before production deployment and repeated when important changes are introduced. A model that performs well on a small or unrealistic test set may not perform equally well with real-world inputs.

Question 115

Which Azure AI Speech capability converts one spoken language into another?

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

Correct Answer: 1

Explanation

Speech translation can process spoken language and produce a translation into another supported language. This capability can be useful for multilingual meetings, customer interactions, travel applications, and communication systems. It differs from standard speech-to-text, which primarily converts speech into written language without necessarily translating it. Developers should consider language support, audio quality, latency, and accuracy requirements when designing a speech translation solution. Testing should use realistic accents, environments, vocabulary, and conversation patterns expected in the application.

Question 116

A RAG application retrieves too many irrelevant documents. Which change could improve retrieval quality?

1. Increase unrelated documents
2. Improve chunking and search configuration
3. Remove all metadata
4. Disable retrieval

Correct Answer: 2

Explanation

Retrieving too many irrelevant documents can reduce the quality of the context provided to a language model. Developers can investigate chunk size, overlap, metadata filters, query formulation, search parameters, vector similarity, semantic ranking, and reranking. The goal is to retrieve a smaller set of highly relevant information rather than overwhelming the model with unrelated content. Evaluation using representative queries can help identify which retrieval changes are effective. Retrieval quality should be measured before and after configuration changes rather than assumed to improve automatically.

Question 117

Which practice helps prevent an AI application from exposing unnecessary personal data?

1. Data minimization
2. Public data sharing
3. Unlimited logging
4. Administrator access for all users

Correct Answer: 1

Explanation

Data minimization reduces privacy exposure by limiting the collection, processing, and retention of personal information to what is necessary for the application’s purpose. For AI applications, this can include removing unnecessary personal details before sending content to a model or storing information in logs. Developers should also use appropriate access controls, encryption, retention policies, and secure storage. Data minimization does not eliminate all privacy risks, but it reduces the amount of sensitive information that needs to be handled by the system.

Question 118

What is the main benefit of using metadata filters during retrieval?

1. They convert text to speech
2. They restrict results based on attributes
3. They train a model automatically
4. They generate images

Correct Answer: 2

Explanation

Metadata filters allow a search application to restrict retrieval results based on attributes associated with indexed content. Metadata might include department, document type, date, region, access level, or other properties. For example, an application could retrieve only documents belonging to a particular department or documents updated within a specific period. Filters can improve relevance and can also support access-control requirements when implemented correctly. Developers should ensure that security-sensitive filters cannot be bypassed through user queries or application logic.

Question 119

Which approach is useful for reducing the risk of prompt injection?

1. Trust every user instruction
2. Ignore retrieved content boundaries
3. Separate trusted instructions from untrusted content
4. Give the model administrator permissions

Correct Answer: 3

Explanation

Prompt injection occurs when untrusted content attempts to influence an AI system’s behavior or override intended instructions. Developers can reduce risk by clearly separating trusted instructions from untrusted user or retrieved content, limiting tool permissions, validating inputs and outputs, and applying appropriate security controls. Applications should not assume that retrieved documents are trustworthy merely because they came from a search system. Sensitive actions should require authorization and, when appropriate, human approval. Prompt defenses should be combined with application-level security rather than relying on instructions alone.

Question 120

Why should AI applications validate model-generated tool arguments before execution?

1. To improve screen resolution
2. To prevent invalid or unsafe operations
3. To create more embeddings
4. To increase audio volume

Correct Answer:2

Explanation

Tool argument validation helps ensure that values generated by a language model meet the application’s expected format, type, range, and business rules before an external action is performed. This is important because model-generated arguments are not automatically trustworthy. For example, an application should verify an account identifier, amount, or requested operation before sending it to a business API. Validation can prevent malformed requests and reduce the risk of unintended actions. Sensitive operations should also use authorization checks and appropriate approval workflows.