AI-901 Practice Test 1 – 50 Questions and Answers (Azure AI Fundamentals)

Published on

AI-901 Practice Test 1: 50 Questions and Answers for Azure AI Fundamentals

Practice AI-901 questions covering responsible AI, generative AI, Microsoft Foundry, AI agents, text and speech solutions, computer vision, image generation, and information extraction. Select your answers, review detailed explanations, and track your progress in the Learning Dashboard.

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

Before you start

This free 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.

The questions are original educational practice questions for Microsoft Azure AI Fundamentals learners. This site is independent and is not affiliated with Microsoft.

What this AI-901 practice test covers

The test is aligned with the current AI-901 direction: identifying AI concepts and capabilities, then implementing AI solutions by using Microsoft Foundry. Topics include responsible AI, model selection, prompts, deployed models, AI agents, text analysis, speech, vision, image generation, retrieval-augmented generation, and Azure Content Understanding.

AI-901 practice test questions

Question 1: Which principle of responsible AI focuses on ensuring that an AI system does not unfairly advantage or disadvantage groups of users?

The correct answer is Fairness.

Fairness means an AI solution should be designed and evaluated to reduce unfair bias and avoid discriminatory outcomes. In practice, teams review data, test model behavior across user groups, and monitor the system after deployment. Fairness is especially important when AI influences decisions about people, access, or recommendations.

Related Microsoft Learn topic

Identify principles and practices for responsible AI

Question 2: Transparency in responsible AI means users should be able to understand when they are interacting with AI and how important AI-assisted decisions are made.

The correct answer is true.

Transparency requires clear communication about AI usage, system limitations, and the factors that influence outputs. It helps users make informed decisions and supports trust in AI-enabled applications. Transparent systems are easier to audit, document, and improve.

Related Microsoft Learn topic

Identify principles and practices for responsible AI

Question 3: Which considerations are part of responsible AI. Choose 3 answers.

The correct answers are Privacy and security, Inclusiveness, and Accountability.

Responsible AI includes privacy and security, inclusiveness, accountability, fairness, transparency, and reliability and safety. These principles help teams design AI systems that protect users, work for diverse populations, and have clear ownership. They are governance principles, not infrastructure maintenance tasks.

Related Microsoft Learn topic

Identify principles and practices for responsible AI

Question 4: Which term best describes an AI model that creates new text, images, code, or other content based on patterns learned from training data?

The correct answer is Generative AI model.

Generative AI models produce new content rather than only predicting a label or numeric value. Large language models can generate text, summarize content, write code, and support conversational experiences. Image-generation models create visual outputs from prompts or other inputs.

Related Microsoft Learn topic

Azure OpenAI in Azure AI Foundry Models

Question 5: A company wants to automatically group customer support tickets by topic without predefined labels. Which machine learning approach is most appropriate?

The correct answer is Clustering.

Clustering is an unsupervised learning technique that groups similar items without requiring predefined categories. It can help discover themes or natural groupings in support tickets. Classification would be more appropriate when the categories are already known and labeled training data exists.

Related Microsoft Learn topic

Fundamental machine learning concepts

Question 6: Which workload is used to detect objects, read text from images, or analyze visual content?

The correct answer is Computer vision.

Computer vision enables applications to analyze images and video. Common scenarios include optical character recognition, object detection, image classification, face detection, and visual question answering. It is the correct workload for understanding visual input.

Related Microsoft Learn topic

Computer Vision overview

Question 7: Sentiment analysis is primarily used to convert spoken audio into written text.

The correct answer is false.

Sentiment analysis evaluates text to determine whether the expressed opinion is positive, negative, neutral, or mixed. Converting spoken audio into written text is speech recognition, also known as speech-to-text. These are different AI workloads.

Related Microsoft Learn topic

Sentiment analysis and opinion mining

Question 8: Which NLP capability identifies names, places, dates, quantities, and organizations in text?

The correct answer is Named entity recognition.

Named entity recognition detects and categorizes entities in unstructured text. It can identify people, organizations, locations, dates, quantities, and other known entity types. This helps applications extract structured information from documents and messages.

Related Microsoft Learn topic

Named entity recognition

Question 9: Which capabilities are common text analysis techniques. Choose 3 answers.

The correct answers are Key phrase extraction, Entity recognition, and Summarization.

Text analysis commonly includes extracting key phrases, detecting entities, analyzing sentiment, identifying language, and summarizing content. Image generation and speech synthesis are AI capabilities, but they are not text analysis techniques.

Related Microsoft Learn topic

Azure AI Language overview

Question 10: Which capability converts written text into spoken audio?

The correct answer is Text-to-speech.

Text-to-speech converts text into synthesized spoken audio. It is commonly used in accessibility tools, voice assistants, notification systems, and conversational applications. Speech-to-text performs the opposite operation by transcribing audio into text.

Related Microsoft Learn topic

Text to speech overview

Question 11: Which capability converts spoken audio into written text?

The correct answer is Speech-to-text.

Speech-to-text transcribes spoken audio into written text. It can be used for captions, call center analytics, dictation, meeting transcription, and voice-enabled applications. It is part of Azure AI Speech capabilities.

Related Microsoft Learn topic

Speech to text overview

Question 12: A multimodal model can process more than one type of input, such as text and images.

The correct answer is true.

Multimodal models can work with multiple input or output types, such as text, images, audio, or video. For example, a model might answer questions about an image or generate text based on visual input. Multimodal capabilities are important for modern generative AI scenarios.

Related Microsoft Learn topic

Azure OpenAI models

Question 13: Which AI workload is best suited for extracting fields such as invoice number, date, vendor, and total amount from scanned invoices?

The correct answer is Information extraction.

Information extraction identifies structured data inside unstructured or semi-structured content. Invoices, receipts, identity documents, forms, and contracts are common examples. Azure Content Understanding and related AI services can help extract fields from documents and other media.

Related Microsoft Learn topic

Azure AI Content Understanding overview

Question 14: Which scenarios are examples of agentic AI. Choose 2 answers.

The correct answers are An AI assistant that uses tools to complete a multi-step user request and An AI agent that plans steps, calls APIs, and returns a final answer.

Agentic AI systems can reason over a goal, plan actions, use tools or APIs, and produce a response based on intermediate steps. They are different from static dashboards or infrastructure automation that does not use AI reasoning. Agentic patterns are increasingly important in AI-901 because they connect generative AI with practical task completion.

Related Microsoft Learn topic

Azure AI Foundry Agent Service

Question 15: Which model type is typically used to predict a numeric value, such as the expected price of a house?

The correct answer is Regression.

Regression predicts continuous numeric values. Examples include price prediction, demand forecasting, temperature prediction, and estimating delivery time. Classification predicts discrete categories rather than numeric values.

Related Microsoft Learn topic

Fundamental machine learning concepts

Question 16: Which model type is used to assign an input to one of several predefined categories?

The correct answer is Classification.

Classification models predict a class or label from predefined categories. Examples include spam detection, image category prediction, and routing support tickets to known departments. If there are only two possible categories, it is binary classification.

Related Microsoft Learn topic

Fundamental machine learning concepts

Question 17: Reliability and safety in responsible AI means an AI system should always provide confident answers even when it lacks enough context.

The correct answer is false.

Reliable and safe AI systems should handle uncertainty, avoid harmful behavior, and communicate limitations when appropriate. A system that confidently invents answers can create risk for users and organizations. Guardrails, testing, monitoring, and fallback behavior help improve reliability and safety.

Related Microsoft Learn topic

Identify principles and practices for responsible AI

Question 18: Which responsible AI principle requires that humans or organizations remain answerable for AI system decisions and outcomes?

The correct answer is Accountability.

Accountability means people and organizations must be responsible for AI systems they design, deploy, and operate. It includes governance, documentation, ownership, escalation paths, and review processes. AI systems should not remove responsibility from the people using or managing them.

Related Microsoft Learn topic

Identify principles and practices for responsible AI

Question 19: Which AI capability is most appropriate for generating a short summary of a long customer email?

The correct answer is Summarization.

Summarization condenses long text into a shorter version while preserving the key meaning. It is useful for emails, support cases, articles, transcripts, and reports. Generative AI models can also perform abstractive summarization with natural language output.

Related Microsoft Learn topic

Summarization overview

Question 20: Which features are commonly associated with computer vision and image-generation workloads. Choose 2 answers.

The correct answers are Detecting objects in images and Creating a new image from a prompt.

Computer vision includes analyzing visual input, such as detecting objects or reading text in images. Image-generation models create new visual outputs based on prompts or other inputs. Numeric prediction, speech transcription, and access control belong to other domains.

Related Microsoft Learn topic

Computer Vision overview

Question 21: Which concept describes the instructions and context sent to a generative AI model to guide its response?

The correct answer is Prompt.

A prompt is the input instruction, question, context, or example provided to a generative AI model. Prompt design affects the quality, safety, and relevance of responses. AI-901 expects foundational knowledge of system prompts and user prompts for generative AI applications.

Related Microsoft Learn topic

Prompt engineering concepts

Question 22: Privacy and security considerations are important when sending user data to an AI model or storing AI-generated outputs.

The correct answer is true.

AI applications often process sensitive user data, documents, prompts, or generated responses. Teams should apply privacy, security, data retention, encryption, access control, and compliance controls. Responsible AI requires protecting both input data and AI outputs.

Related Microsoft Learn topic

Identify principles and practices for responsible AI

Question 23: Which Azure service provides a unified platform for building, evaluating, deploying, and managing AI applications and agents?

The correct answer is Microsoft Foundry.

Microsoft Foundry provides tools for working with AI models, prompts, agents, evaluations, and application integration. It is the main platform emphasized in AI-901 for implementing AI solutions on Azure. It helps developers move from experimentation to deployed AI applications.

Related Microsoft Learn topic

What is Azure AI Foundry?

Question 24: In a generative AI chat application, which prompt type is commonly used to define the assistant's behavior, rules, and boundaries?

The correct answer is System prompt.

A system prompt defines high-level instructions for how the model should behave. It can specify tone, safety rules, response format, and constraints. A user prompt is the specific request submitted by the user during the interaction.

Related Microsoft Learn topic

Prompt engineering concepts

Question 25: Which actions are part of implementing a generative AI app in Microsoft Foundry. Choose 3 answers.

The correct answers are Deploying a model, Testing prompts, and Calling the model from a lightweight client application.

A typical Foundry workflow includes choosing and deploying a model, designing prompts, testing behavior, and calling the deployed model from an application. Storage tiers and VPN gateways may be part of broader Azure architecture, but they are not the core AI app implementation tasks measured here.

Related Microsoft Learn topic

What is Azure AI Foundry?

Question 26: A deployed model endpoint can be called from an application by using an SDK, REST API, or supported client library.

The correct answer is true.

After a model is deployed, applications can send requests to the endpoint using supported APIs, SDKs, or client libraries. This enables integration with chat clients, web applications, automation tools, and backend services. Developers must also handle authentication and configuration securely.

Related Microsoft Learn topic

Azure AI Foundry SDK overview

Question 27: Which Foundry capability helps create an AI system that can use tools and perform multi-step tasks on behalf of a user?

The correct answer is Agent service.

Agent capabilities in Foundry help create AI assistants that can use tools, follow instructions, and complete multi-step tasks. Agents can be tested in the portal and integrated into lightweight client applications. This supports agentic AI scenarios such as research assistants, support bots, and workflow copilots.

Related Microsoft Learn topic

Azure AI Foundry Agent Service

Question 28: A developer wants a chat app to answer questions about an uploaded image. Which model capability is required?

The correct answer is Multimodal vision input.

The application needs a multimodal model that can interpret visual input in prompts. A text-only model cannot directly analyze the image unless another service first extracts information from it. Multimodal models are commonly used for visual question answering and image interpretation scenarios.

Related Microsoft Learn topic

Azure OpenAI models

Question 29: Which Foundry Tools capability is designed to extract structured information from documents, images, audio, and video?

The correct answer is Azure Content Understanding.

Azure Content Understanding extracts information from multiple content types, including documents, images, audio, and video. It can support scenarios such as invoice processing, form extraction, media analysis, and structured data extraction from unstructured content. AI-901 includes information extraction by using Content Understanding in Foundry Tools.

Related Microsoft Learn topic

Azure AI Content Understanding overview

Question 30: Which scenarios can be supported by Azure Content Understanding. Choose 3 answers.

The correct answers are Extracting fields from a form, Extracting information from an image, and Extracting information from audio or video.

Content Understanding is used for information extraction across content types, including forms, documents, images, audio, and video. Network traffic distribution and RBAC assignment are Azure management tasks, not Content Understanding scenarios.

Related Microsoft Learn topic

Azure AI Content Understanding overview

Question 31: A developer wants to build an application that analyzes customer reviews for sentiment and key phrases. Which Foundry-related AI capability should be used?

The correct answer is Azure AI Language.

Azure AI Language provides text analysis capabilities such as sentiment analysis, key phrase extraction, language detection, and entity recognition. These services can be used in lightweight applications that process customer feedback. Speech and networking services do not provide the required text analysis capabilities.

Related Microsoft Learn topic

Azure AI Language overview

Question 32: To use a generative AI model in an application, you must always train a new model from scratch.

The correct answer is false.

Most generative AI applications use existing foundation models and configure them with prompts, parameters, grounding data, or fine-tuning when appropriate. Training a new model from scratch is expensive and usually unnecessary for fundamentals-level application scenarios. Foundry focuses on selecting, deploying, testing, and integrating models.

Related Microsoft Learn topic

What is Azure AI Foundry?

Question 33: Which parameter commonly controls how creative or random a generative AI model's responses are?

The correct answer is Temperature.

Temperature influences randomness in model output. Lower values usually make responses more focused and deterministic, while higher values can increase variety and creativity. It is an important model configuration parameter for generative AI applications.

Related Microsoft Learn topic

Azure OpenAI reference

Question 34: Which approach helps reduce irrelevant or invented responses by providing trusted source content to a generative AI app?

The correct answer is Grounding the model with relevant data.

Grounding provides relevant context or source data to help the model generate more accurate responses. Retrieval-augmented generation is a common grounding pattern that retrieves documents or passages before prompting the model. This can reduce hallucinations and improve traceability.

Related Microsoft Learn topic

Retrieval augmented generation

Question 35: Which practices help improve prompt quality for a generative AI application. Choose 2 answers.

The correct answers are Give clear instructions and expected output format and Provide relevant examples or constraints.

Effective prompts include clear instructions, context, constraints, and sometimes examples. Defining the expected output format helps the application parse and display responses consistently. Prompts should never expose secrets or ignore safety and reliability constraints.

Related Microsoft Learn topic

Prompt engineering concepts

Question 36: Which capability should a Foundry-based app use to respond to spoken prompts with generated answers?

The correct answer is Speech and multimodal AI capabilities.

An application that responds to spoken prompts needs speech input processing and an AI model that can generate a response. Depending on the design, it may use speech-to-text, a multimodal model, and text-to-speech. Policy, networking, and storage lifecycle features do not provide conversational speech behavior.

Related Microsoft Learn topic

Azure AI Speech overview

Question 37: Azure AI Speech can be used by applications that need speech recognition or speech synthesis capabilities.

The correct answer is true.

Azure AI Speech supports speech-to-text and text-to-speech scenarios. These capabilities are useful for voice assistants, accessibility, call transcription, captions, and spoken responses. Foundry-based applications can combine speech with other AI capabilities for richer user experiences.

Related Microsoft Learn topic

Azure AI Speech overview

Question 38: A team wants to create a lightweight client application that calls an AI agent. What must the application typically include?

The correct answer is Endpoint configuration and authentication to call the agent.

A client application must know how to connect to the agent or model endpoint and authenticate securely. It also needs code to send user messages, receive responses, and handle errors. The application does not require physical servers or public IP addresses for every user.

Related Microsoft Learn topic

Azure AI Foundry Agent Service

Question 39: Which Azure AI capability is best for creating a new image from a natural language prompt?

The correct answer is Image generation.

Image generation models create new visual content based on prompts or other inputs. They are used for design concepts, illustrations, creative content, and synthetic visual generation scenarios. This is different from computer vision analysis, which interprets existing images.

Related Microsoft Learn topic

Azure OpenAI models

Question 40: Which tasks are relevant when testing a single-agent solution in the Foundry portal. Choose 3 answers.

The correct answers are Check whether the agent follows instructions, Test tool usage with realistic prompts, and Validate responses against expected outcomes.

Agent testing should verify instruction following, tool usage, response quality, and expected outcomes. Realistic test prompts help reveal gaps before integration with a client app. Removing logs or disabling safety controls is not a good testing practice.

Related Microsoft Learn topic

Azure AI Foundry Agent Service

Question 41: Which practice helps protect secrets used by an AI client application?

The correct answer is Store secrets in a secure configuration service such as Azure Key Vault.

Secrets such as API keys and connection strings should be stored securely and accessed through controlled mechanisms. Azure Key Vault helps protect keys, secrets, and certificates. Prompts, public repositories, and chat responses must not expose sensitive credentials.

Related Microsoft Learn topic

Azure Key Vault overview

Question 42: When building AI applications, developers should validate and handle model responses instead of assuming every response is perfectly formatted.

The correct answer is true.

AI model responses can vary, so applications should validate outputs, handle errors, and apply fallback logic when needed. This is especially important when responses are parsed as JSON, shown to users, or used in workflows. Validation improves reliability and safety.

Related Microsoft Learn topic

Evaluation and monitoring for generative AI

Question 43: Which capability is most useful when an application needs to answer questions based on a private knowledge base?

The correct answer is Retrieval-augmented generation.

Retrieval-augmented generation retrieves relevant content from a knowledge base and supplies it to the model as context. This helps the model answer using organization-specific data without retraining the foundation model. It also helps provide source-grounded responses.

Related Microsoft Learn topic

Retrieval augmented generation

Question 44: Which output format instruction is most helpful when an application needs to process the model response programmatically?

The correct answer is Return valid JSON with specific fields.

When an application needs to parse a response, the prompt should clearly specify the required structure. Asking for valid JSON with known fields can make downstream processing more reliable. The application should still validate the response before using it.

Related Microsoft Learn topic

Prompt engineering concepts

Question 45: Which tasks are common when implementing text and speech AI solutions by using Foundry. Choose 3 answers.

The correct answers are Build a lightweight application that analyzes text, Use speech capabilities for spoken prompts, and Use Azure AI Speech in Foundry Tools.

AI-901 includes building lightweight applications that use text analysis and speech capabilities. Foundry Tools can help integrate Azure AI Speech for voice-enabled scenarios. Router configuration and disk resizing are infrastructure tasks outside this objective.

Related Microsoft Learn topic

Azure AI Speech overview

Question 46: A developer wants an AI app to interpret a product photo and answer questions about what is visible. Which input should the model be able to process?

The correct answer is Visual input in prompts.

The model must support visual input to interpret a product photo. Multimodal models can process images along with text instructions, enabling visual question answering. Without visual input support, the app would need another service to extract image details first.

Related Microsoft Learn topic

Azure OpenAI models

Question 47: Which capability helps developers compare model outputs against quality, safety, or task-specific criteria?

The correct answer is Evaluation.

Evaluation helps measure whether an AI solution produces useful, safe, and relevant outputs. Developers can test prompts, compare responses, and identify regressions before production release. Evaluation is a key part of building reliable AI applications.

Related Microsoft Learn topic

Evaluation and monitoring for generative AI

Question 48: A Foundry-based AI app should expose all raw system prompts to every end user so users can change the application's safety rules.

The correct answer is false.

System prompts and safety instructions define how the application should behave and should be controlled by the application owner. Users can provide task-specific prompts, but they should not be allowed to override security, privacy, or safety boundaries. Prompt injection defenses are important for production AI apps.

Related Microsoft Learn topic

Prompt engineering concepts

Question 49: Which AI service capability is best suited for extracting text from an image of a receipt before analyzing the receipt contents?

The correct answer is Optical character recognition.

Optical character recognition detects and extracts printed or handwritten text from images and documents. It is commonly used in receipt processing, document digitization, and search indexing workflows. After OCR, the extracted text can be analyzed or passed to other AI services.

Related Microsoft Learn topic

Optical character recognition overview

Question 50: Which choices are appropriate model selection considerations. Choose 2 answers.

The correct answers are Whether the model supports the required input and output modalities and Latency, cost, and quality requirements.

Model selection should consider whether the model can perform the required task, support needed modalities, meet latency and cost requirements, and produce acceptable quality. Privacy, testing, and security requirements should never be ignored. Workstation monitor count is irrelevant.

Related Microsoft Learn topic

Azure OpenAI models

Comments

Post a Comment