AI-901 Practice Test 2 – 50 Questions and Answers (Azure AI Fundamentals)

Published on

AI-901 Practice Test 2: 50 Questions and Answers for Azure AI Fundamentals

Practice AI-901 questions for Microsoft Azure AI Fundamentals. This free test covers responsible AI, generative AI, AI agents, Microsoft Foundry, model deployment, prompts, Azure Language, Azure Speech, Azure Vision, and Content Understanding. Select your answers, review the explanations, and track your progress in the Learning Dashboard.

Exam: AI-901Questions: 50Recommended score: 70%+Time: 60 minutes

Before you start

This AI-901 practice test includes single-choice, multiple-response, and true/false questions. When a question requires more than one answer, the question text tells you exactly how many answers to choose.

This independent educational practice test is designed to help learners prepare for Azure AI Fundamentals. It is not a Microsoft exam dump and does not contain real exam questions.

AI-901 practice test questions

Question 1: A bank uses an AI model to approve or reject loan applications. Which responsible AI principle is most directly related to checking whether the model treats different demographic groups consistently?

The correct answer is Fairness.

Fairness focuses on reducing unjustified differences in outcomes across people and groups. In this scenario, the organization should evaluate whether the model creates biased approval patterns and adjust the solution before production use.

Related Microsoft Learn topic

AI-901 study guide

Question 2: Which responsible AI principle focuses on designing AI systems that behave dependably and avoid harmful outcomes when conditions change?

The correct answer is Reliability and safety.

Reliability and safety means the AI solution should work as intended and handle errors or unusual inputs appropriately. It is especially important for systems used in healthcare, finance, transportation, or other high-impact environments.

Related Microsoft Learn topic

AI-901 study guide

Question 3: An AI application processes customer support messages that contain names, phone numbers, and account IDs. Which responsible AI area should be prioritized before sending this data to a model?

The correct answer is Privacy and security.

Privacy and security require protecting personal and sensitive data throughout the AI workflow. The application should minimize unnecessary data exposure, use appropriate access controls, and handle personally identifiable information carefully.

Related Microsoft Learn topic

AI-901 study guide

Question 4: Transparency in responsible AI can include explaining the purpose, limitations, and expected behavior of an AI system to users.

The correct answer is true.

Transparency helps people understand how and why an AI system is used. Clear communication about capabilities and limitations supports trust and helps users make informed decisions.

Related Microsoft Learn topic

AI-901 study guide

Question 5: A company assigns a named product owner to review model risks, approve release decisions, and respond when users report AI issues. Which responsible AI principle does this support?

The correct answer is Accountability.

Accountability means people and organizations remain responsible for how AI systems are designed, deployed, and operated. Assigning ownership ensures there is a clear process for governance, review, and remediation.

Related Microsoft Learn topic

AI-901 study guide

Question 6: Which type of AI model is commonly used to generate natural-language answers, summaries, and text completions from prompts?

The correct answer is Large language model.

Large language models are trained on large amounts of text and can generate, transform, and analyze language. They are commonly used for chat, summarization, content generation, and question-answering scenarios.

Related Microsoft Learn topic

Introduction to generative AI and agents

Question 7: Which configuration choices can affect the behavior of a deployed generative AI model. Choose 2 answers.

The correct answers are Temperature and Maximum output tokens.

Temperature influences how deterministic or creative model responses may be, while maximum output tokens limits response length. These parameters help tune model behavior for different application requirements.

Related Microsoft Learn topic

AI-901 study guide

Question 8: You want a generative AI response to be more predictable and less creative. Which parameter adjustment is most appropriate?

The correct answer is Use a lower temperature value.

A lower temperature generally makes output more focused and deterministic. This is useful when the application needs consistent answers, structured responses, or limited variation.

Related Microsoft Learn topic

Introduction to generative AI and agents

Question 9: Which AI workload is best suited for identifying whether a product review expresses positive, negative, or neutral feedback?

The correct answer is Sentiment analysis.

Sentiment analysis evaluates text to estimate the attitude or emotion expressed by the writer. It is commonly used for customer reviews, support messages, survey responses, and social feedback.

Related Microsoft Learn topic

Sentiment analysis and opinion mining

Question 10: Generative AI can create new content such as text, images, code, summaries, and conversational responses.

The correct answer is true.

Generative AI systems produce new outputs based on prompts, context, and model training. Common examples include writing responses, summarizing documents, generating images, and assisting with code.

Related Microsoft Learn topic

Introduction to generative AI and agents

Question 11: A support team wants to automatically identify the main topics in thousands of customer comments. Which NLP capability should they use?

The correct answer is Key phrase extraction.

Key phrase extraction identifies important concepts and topics in unstructured text. It helps teams quickly understand large volumes of comments without reading each message manually.

Related Microsoft Learn topic

Key phrase extraction

Question 12: Which NLP capability identifies people, locations, organizations, dates, and other named items in text?

The correct answer is Named entity recognition.

Named entity recognition extracts known entity types from text, such as names, places, organizations, and dates. This is useful for indexing, enrichment, compliance workflows, and information retrieval.

Related Microsoft Learn topic

Azure Language overview

Question 13: Which capabilities are common natural language processing workloads. Choose 3 answers.

The correct answers are Summarization, Entity detection, and Language detection.

NLP workloads analyze or generate language-based content. Summarization, entity detection, and language detection are common examples used to process text at scale.

Related Microsoft Learn topic

Azure Language overview

Question 14: Which speech capability converts spoken audio into written text?

The correct answer is Speech to text.

Speech to text transcribes audio into text and can support real-time or batch transcription scenarios. It is used for captions, meeting notes, call center analytics, and voice-enabled applications.

Related Microsoft Learn topic

Speech to text

Question 15: Which speech capability creates natural-sounding spoken audio from written text?

The correct answer is Text to speech.

Text to speech converts written text into synthesized audio. It is useful for accessibility, virtual assistants, audio content, and conversational AI experiences.

Related Microsoft Learn topic

Text to speech

Question 16: Language identification can be useful when an application receives audio or text in more than one possible language.

The correct answer is true.

Language identification helps determine the language being used before selecting the right model, voice, or translation path. This is important for multilingual applications and global user experiences.

Related Microsoft Learn topic

Language and voice support

Question 17: Which computer vision capability identifies objects in an image and returns their locations with bounding boxes?

The correct answer is Object detection.

Object detection finds objects in images and returns their locations, often as bounding boxes. This enables applications to count, locate, or react to objects in visual content.

Related Microsoft Learn topic

Object detection

Question 18: Which scenarios are examples of computer vision workloads. Choose 3 answers.

The correct answers are Reading text from a scanned invoice, Detecting a damaged product in a photo, and Generating a caption for an image.

Computer vision workloads interpret visual information from images or documents. OCR, defect detection, and image captioning are common examples of vision-based AI scenarios.

Related Microsoft Learn topic

Azure Vision overview

Question 19: Which technology extracts printed or handwritten text from images and scanned documents?

The correct answer is Optical character recognition.

Optical character recognition, or OCR, extracts readable text from images and documents. It is useful for digitizing forms, receipts, reports, product labels, and handwritten content.

Related Microsoft Learn topic

OCR overview

Question 20: A company needs to extract structured fields from documents, images, audio, and video files. Which capability best matches this requirement?

The correct answer is Content Understanding.

Content Understanding is designed to analyze different media types and transform content into structured output. It can support document, image, audio, and video extraction scenarios.

Related Microsoft Learn topic

Content Understanding overview

Question 21: An AI agent can use instructions, context, and tools to help complete tasks beyond a single text response.

The correct answer is true.

Agentic AI solutions can combine a model with instructions, tools, knowledge, and orchestration to complete tasks. This differs from a simple one-turn prompt because the agent can use additional capabilities during the workflow.

Related Microsoft Learn topic

Introduction to generative AI and agents

Question 22: In a generative AI app, which prompt type is best for defining the assistant's role, tone, boundaries, and behavioral rules?

The correct answer is System prompt.

A system prompt sets high-level instructions for how the model should behave. It can define tone, constraints, safety boundaries, formatting expectations, and the purpose of the assistant.

Related Microsoft Learn topic

Introduction to generative AI and agents

Question 23: A user asks a chatbot, 'Summarize this document in five bullet points.' What kind of prompt is this?

The correct answer is User prompt.

A user prompt is the instruction or request provided by the person interacting with the application. The system prompt may define the assistant's behavior, but the user prompt provides the task for that interaction.

Related Microsoft Learn topic

Introduction to generative AI and agents

Question 24: A lightweight client application that calls a deployed model typically needs connection information such as an endpoint and credentials.

The correct answer is true.

Client applications need a way to reach the deployed model and authenticate requests. In production, credentials should be stored securely and not hardcoded into public client-side code.

Related Microsoft Learn topic

Microsoft Foundry documentation

Question 25: You want to test a generative model interactively before writing application code. Where should you start in Microsoft Foundry?

The correct answer is The model playground or Foundry portal testing experience.

Microsoft Foundry provides portal experiences for selecting, deploying, and testing models before integrating them into code. This helps developers compare behavior, adjust prompts, and validate basic responses quickly.

Related Microsoft Learn topic

Microsoft Foundry documentation

Question 26: A developer is building a chat client that sends prompts to a deployed model from Python code. Which component is most directly used by the application?

The correct answer is A Foundry SDK or REST API client.

A lightweight chat application interacts with the deployed model by using an SDK or REST API. This allows the application to send messages, receive responses, and integrate AI behavior into the user experience.

Related Microsoft Learn topic

Microsoft Foundry documentation

Question 27: Which steps are commonly part of building a lightweight generative AI chat application. Choose 2 answers.

The correct answers are Deploy or select a model endpoint and Send user messages to the model through an SDK or API.

A chat application needs a model endpoint and code that sends prompts and receives responses. Authentication and secret handling should be implemented securely, not bypassed or exposed.

Related Microsoft Learn topic

Microsoft Foundry documentation

Question 28: A support chatbot must always respond with a short answer and a link to documentation. Which prompt-engineering approach is most appropriate?

The correct answer is Specify the required response format in the system instructions.

Clear formatting instructions help the model produce consistent output. For applications that require short responses and documentation links, those expectations should be explicitly stated in the system or developer instructions.

Related Microsoft Learn topic

Introduction to generative AI and agents

Question 29: In Microsoft Foundry, what is the main purpose of creating a single-agent solution?

The correct answer is To configure an AI assistant that can follow instructions and use tools for a task.

A single-agent solution defines an AI assistant with instructions, a model, and optional tools or knowledge. It can be tested and then integrated into applications that need task-oriented AI behavior.

Related Microsoft Learn topic

Microsoft Foundry documentation

Question 30: When creating an agent, clear instructions help define what the agent should do and what it should avoid.

The correct answer is true.

Agent instructions guide the behavior of the model during task execution. Good instructions reduce ambiguity, improve consistency, and help enforce business or safety requirements.

Related Microsoft Learn topic

Introduction to generative AI and agents

Question 31: An agent needs to look up order status from an external system. What should be added to the agent so it can perform that action?

The correct answer is A tool or function that can query the order system.

Tools allow agents to interact with external systems or perform specific actions. Without a tool or integration, the model can only generate a response based on available context and cannot actually query the order system.

Related Microsoft Learn topic

Microsoft Foundry documentation

Question 32: An agent must answer questions using a company's policy documents. Which addition is most useful?

The correct answer is A connected knowledge source or retrieval capability.

A knowledge source or retrieval capability provides relevant grounding information for answers. This helps the agent respond using company-approved content instead of relying only on general model knowledge.

Related Microsoft Learn topic

Microsoft Foundry documentation

Question 33: Which elements are commonly configured when creating an AI agent. Choose 3 answers.

The correct answers are Model, Instructions, and Tools or knowledge sources.

An agent typically needs a model, instructions that define behavior, and optional tools or knowledge sources to support task completion. Physical hardware and printer drivers are not core agent configuration elements in Microsoft Foundry.

Related Microsoft Learn topic

Microsoft Foundry documentation

Question 34: You are building an app that transcribes live meeting audio into captions. Which Foundry Tool capability should you use?

The correct answer is Speech to text.

Speech to text converts spoken audio into written text and supports real-time transcription scenarios. This makes it appropriate for live captions, meeting transcription, and voice-driven interfaces.

Related Microsoft Learn topic

Speech to text

Question 35: A user sends an image and asks, 'What safety issues are visible in this warehouse photo?' Which model capability is required?

The correct answer is Multimodal visual understanding.

A multimodal model can interpret visual input together with a text prompt. This is required when the application needs to reason over images rather than only process text.

Related Microsoft Learn topic

AI-901 study guide

Question 36: Image-generation models can create new visual outputs from text prompts or other supported inputs.

The correct answer is true.

Image-generation models are generative models that create new visual content. They can be used for design prototypes, illustrations, visual concepts, and other scenarios where generated images are needed.

Related Microsoft Learn topic

AI-901 study guide

Question 37: In Content Understanding, what is an analyzer used for?

The correct answer is To define how content is processed and structured output is extracted.

A Content Understanding analyzer defines the extraction task and output structure for the content being processed. It helps transform documents, images, audio, or video into structured information.

Related Microsoft Learn topic

Content Understanding analyzer

Question 38: You need to extract invoice number, vendor name, date, and total amount from uploaded invoices. Which Foundry capability is the best fit?

The correct answer is Content Understanding.

Content Understanding can extract structured fields from documents and forms. It is a strong fit for invoice processing because the required result is a set of fields from business documents.

Related Microsoft Learn topic

Content Understanding overview

Question 39: Which content types can Content Understanding process for information extraction scenarios. Choose 4 answers.

The correct answers are Documents, Images, Audio, and Video.

Content Understanding supports multiple content types, including documents, images, audio, and video. Virtual networks are infrastructure resources, not media content to be analyzed.

Related Microsoft Learn topic

Content Understanding overview

Question 40: A Python app must detect language, extract key phrases, and identify entities from support tickets. Which service capability should it use?

The correct answer is Azure Language in Foundry Tools.

Azure Language provides NLP features for understanding and analyzing text. It supports workloads such as language detection, key phrase extraction, named entity recognition, and sentiment analysis.

Related Microsoft Learn topic

Azure Language overview

Question 41: Application credentials for calling AI services should be stored securely rather than embedded in public source code.

The correct answer is true.

Secrets should be protected using secure configuration, managed identities, or secret stores where appropriate. Exposing credentials in public code can allow unauthorized access and unexpected costs.

Related Microsoft Learn topic

Microsoft Foundry documentation

Question 42: A chat app should display partial model output while the response is still being generated. Which application behavior is being used?

The correct answer is Streaming responses.

Streaming returns generated content progressively rather than waiting for the full response. This can improve user experience in chat applications because users see output sooner.

Related Microsoft Learn topic

Microsoft Foundry documentation

Question 43: Which prompt practices help produce safer and more reliable responses. Choose 2 answers.

The correct answers are Tell the model what sources or context it should use and Specify the expected output format.

Clear context and output-format instructions reduce ambiguity and improve response quality. Vague or unsafe instructions can increase the chance of irrelevant, inconsistent, or risky output.

Related Microsoft Learn topic

Introduction to generative AI and agents

Question 44: A user provides a photo of a damaged package and asks the AI app to describe the damage. Which prompt type is being used?

The correct answer is Multimodal prompt.

A multimodal prompt includes more than one input type, such as text plus an image. The model must be capable of interpreting visual input to answer the user's question correctly.

Related Microsoft Learn topic

AI-901 study guide

Question 45: Microsoft Foundry can be used to build, deploy, and manage AI apps and agents at scale.

The correct answer is true.

Microsoft Foundry provides a platform for building AI applications and agents, working with models, tools, and enterprise management capabilities. It supports moving from prototype to production with governance and operational features.

Related Microsoft Learn topic

What is Microsoft Foundry?

Question 46: You want an AI app to accept spoken questions and respond with spoken answers. Which capabilities are required. Choose 2 answers.

The correct answers are Speech to text and Text to speech.

Speech to text converts the user's spoken input into text for processing. Text to speech converts the generated response back into spoken audio for the user.

Related Microsoft Learn topic

Azure Speech overview

Question 47: An application must read product labels from photos taken by mobile users. Which capability should be included?

The correct answer is OCR or image text extraction.

OCR extracts text from images, including labels, signs, forms, and scanned documents. This allows the application to convert visual text into machine-readable content.

Related Microsoft Learn topic

OCR overview

Question 48: Which outputs are common goals of an information extraction solution. Choose 2 answers.

The correct answers are Structured fields from a document and Searchable data from unstructured content.

Information extraction transforms unstructured or semi-structured content into structured, usable data. This enables search, automation, validation, and downstream business workflows.

Related Microsoft Learn topic

Content Understanding overview

Question 49: A lightweight client application for an agent should typically include which components. Choose 2 answers.

The correct answers are Code that sends user input to the agent endpoint and Logic to display or process the agent response.

An agent client sends user input to the agent and handles the response returned by the service. Secret handling must be implemented securely and should not rely on public hardcoded credentials.

Related Microsoft Learn topic

Microsoft Foundry documentation

Question 50: Responsible AI considerations should continue during implementation, testing, deployment, and monitoring of AI solutions.

The correct answer is true.

Responsible AI is not only a design-time activity. Teams should continue to evaluate fairness, safety, privacy, transparency, and accountability as the solution is built, deployed, and operated.

Related Microsoft Learn topic

AI-901 study guide

Comments

Post a Comment