AI-901 Sample Questions Answers

The completed sentence is:

Ensuring that human reviewers oversee AI-generated decisions and remain responsible for the final output is an example of the Microsoft responsible AI principle of accountability.

Accountability means people and organizations remain responsible for AI systems and their effects. Human review and responsibility for final decisions are examples of accountability.

The other options are incorrect:

fairness focuses on avoiding bias and treating people fairly.

privacy and security focuses on protecting data and restricting access.

transparency focuses on explaining AI use, capabilities, and limitations.

Statement 1: Azure Content Understanding in Foundry Tools can analyze only PDF documents. = No

Azure Content Understanding is not limited to PDF documents. It can analyze different content types, including documents, forms, images, audio, and video.

Statement 2: Azure Content Understanding in Foundry Tools results are returned in the JSON format. = Yes

Azure Content Understanding returns structured analysis output in JSON format.

Statement 3: Azure Content Understanding in Foundry Tools can extract structured fields from documents and forms. = Yes

Azure Content Understanding can extract structured fields from documents and forms by using analyzers and schemas.

A schema defines which fields to extract when analyzing content.

In Azure Content Understanding, the schema or fieldSchema defines the structured data that the analyzer extracts from content, including field names, types, and extraction behavior. Microsoft documentation states that Content Understanding lets you define a schema to extract, classify, or generate field values from unstructured content.

The other options are incorrect:

A keyword list does not define the complete structured output fields.

OCR-only processing extracts text, but it does not define structured fields.

A synchronous API call describes a request pattern, not the extraction schema.

Azure Content Understanding analysis operations are long-running operations in the Python SDK. Microsoft’s Python SDK documentation states that analysis operations return a poller, and the SDK provides LROPoller types that handle polling automatically when you call .result().

Therefore, the correct workflow is to submit the PDF by calling begin_analyze(), receive a poller, and then call:

result = poller.result()

Option B is incorrect because extraction results are not read from request headers.

Option C is incorrect because the requirement is to use Azure Content Understanding extraction, not build a manual OCR-only mapping pipeline.

Option D is incorrect because the SDK analysis pattern is asynchronous/long-running, not a simple synchronous analyze() call that returns all extracted fields in the same request.

Creating new illustrations based on the text of an article is a content generation task. Because the AI solution must create new visual content from a text prompt or written input, the correct workload is generative AI.

Statement 1: To process prompts that contain images, you must deploy a multimodal generative AI model. = Yes

Image prompts require a vision-enabled or multimodal model. A text-only model cannot interpret image content.

Statement 2: In the Foundry playground, you can provide an image by uploading a local file or by specifying a publicly accessible URL. = Yes

The Foundry playground for vision-capable models supports adding image input to a prompt, including local uploads and image URLs.

Statement 3: A multimodal model can interpret multiple images included in a single prompt. = Yes

Vision-enabled multimodal models can accept image content items in a prompt, and multiple images can be included for the model to analyze together.

AudioOutputConfig(filename= " output.wav " )

The question specifically states the application must save the synthesized audio to a file. In the Azure Speech SDK for Python, speechsdk.audio.AudioOutputConfig(filename= " output.wav " ) directs the synthesizer to write the generated speech output directly to a WAV file on disk — which is exactly the requirement.

Why the other options are wrong:

AudioOutputConfig(stream) — This routes audio output to an in-memory audio stream object, not a file. It is used when you want to process or play the audio programmatically without saving it to disk.

AudioStreamFormat(wave_stream_format=AudioStreamWaveFormat.PCM) — This class defines the format of an audio stream (e.g., PCM encoding, sample rate). It is used when configuring custom audio streams, not when specifying a file output destination. It is not a valid argument for speechsdk.audio. in this context.

The correct and complete line is:

audio_config = speechsdk.audio.AudioOutputConfig(filename= " output.wav " )

Microsoft Foundry Agent Service documentation states that instructions define goals, constraints, and behavior for an agent. Therefore, instructions are used to guide how the generative AI model or agent should respond and behave.

Option A is correct because instructions can define constraints the model must follow.

Option B is correct because instructions can define the agent’s role and behavior.

Options C, D, and E are incorrect because Azure region, model selection, and TPM allocation are configuration or deployment/resource settings, not purposes of prompt instructions.

Microsoft’s Foundry Agent Service documentation states that tool_choice provides deterministic control over tool calling:

auto means the model decides whether to call tools.

required means the model must call one or more tools.

none means the model does not call tools.

Therefore:

A. auto = Incorrect, because the model may or may not call the Azure function.

B. none = Incorrect, because this prevents tool/function calls.

C. required = Correct, because it forces the agent to call a tool.

The Azure OpenAI function-calling documentation also confirms that tool_choice= " auto " lets the model decide whether to call a function, while tool_choice= " none " forces a user-facing response without a tool call.

To return important topics from unstructured text by using Azure Language SDK, use:

client.extract_key_phrases()

Microsoft states that key phrase extraction evaluates unstructured text and returns a list of key phrases or main topics. The Azure Text Analytics client library for Python includes the extract_key_phrases method for this task.

Why the other options are incorrect:

A. client.recognize_linked_entities() identifies entities and links them to a knowledge base.

B. client.recognize_entities() identifies named entities such as people, organizations, locations, dates, and quantities.

C. client.begin_extract_summary() starts an extractive summarization operation, not key phrase extraction.

Capturing or reading text from images is optical character recognition (OCR). OCR is used to detect and extract printed or handwritten text from images, scanned documents, and other visual content.

A system prompt, also called a system message, is used to guide how a generative AI model behaves, including its tone, role, response style, and output format. Microsoft states that system messages help steer an Azure OpenAI chat model toward the desired behavior, tone, and output format.

Statement 1: System prompts can be used to authorize users. = No

Authorization is handled by application identity, access control, authentication, and security configuration. A system prompt can guide model behavior, but it is not an authorization mechanism.

Statement 2: A system prompt is used to reduce tokens per minute (TPM). = No

TPM is a quota or throughput setting for model deployments. A system prompt does not reduce or configure TPM.

Statement 3: A system prompt guides the behavior of a generative AI model. = Yes

This is correct. Microsoft documentation says system messages help steer the model’s behavior, tone, and output format.

client = OpenAI(

api_key= " ... " ,

base_url= " https://resource1.openai.azure.com/openai/v1/ " ,

)

response = client.responses.create(

model= " my-mini-gpt " ,

)

For Azure OpenAI in Microsoft Foundry, the base_url uses the Azure OpenAI resource name in the endpoint format:

https:// < resource-name > .openai.azure.com/openai/v1/

In the question, the Azure OpenAI resource is named Resource1, so the first blank must be resource1. Microsoft documentation for Azure OpenAI v1 endpoints confirms that the endpoint must use the ...openai.azure.com/openai/v1/ path.

For the model parameter, Azure OpenAI requires the deployment name, not the underlying model name. Microsoft states that Azure OpenAI always requires the deployment name when calling APIs, even when the parameter is named model.

The deployed model is gpt-4.1-mini, but the deployment name is my-mini-gpt. Therefore, the second blank must be:

model= " my-mini-gpt "

So the correct selections are:

base_url blank = resource1

model blank = my-mini-gpt

Statement 1: Voice Live returns only transcribed text. = No

Voice Live is not limited to returning transcribed text. It is designed for real-time spoken conversational experiences and can include speech input, model reasoning, and spoken output.

Statement 2: Voice Live requires you to separately implement speech to text and text to speech services. = No

Voice Live combines the conversational voice pipeline, so you do not need to separately build independent speech-to-text and text-to-speech services for the same real-time voice experience.

Statement 3: Voice Live combines speech to text, reasoning, and text to speech into a single conversational experience. = Yes

This statement is correct. Voice Live is used for real-time conversational AI experiences where spoken input is processed, the model reasons over the request, and spoken output can be returned.

The requirement is to extract structured fields, including nested values, from invoices by using a defined schema. In Azure Content Understanding, an analyzer is the processing unit that defines how content is analyzed, what information is extracted, and how the output is structured, including JSON fields.

Microsoft’s Content Understanding document solutions documentation states that Content Understanding uses customizable analyzers to extract essential information, fields, and relationships from documents and forms. Microsoft’s quickstart also shows invoice processing with the prebuilt-invoice analyzer to extract structured data from an invoice document.

Why the other options are incorrect:

A. transcription workflow in Azure Speech is for converting audio to text, not invoice field extraction.

B. OCR-only document processing can extract text but does not meet the requirement for structured fields and nested values by schema.

D. Azure AI Search is for indexing and querying content, not defining invoice extraction schemas.

Therefore, the correct answer is C. an analyzer in Azure Content Understanding in Foundry Tools.

To minimize unintended bias, the training data should be representative of the population for which the model will be used. Microsoft responsible AI guidance identifies fairness as a key principle and notes that discrepancies can occur when systems underperform for specific demographic groups or rarely observed input conditions in training data.

A global insurance dataset may not represent UK drivers. A completely random sample may still underrepresent important groups. Removing protected characteristics alone does not guarantee fairness, because other fields can act as proxies. The best answer is to use a training sample representative of the target UK population.

Use a provisioned deployment when the solution requires a specific allocated amount of throughput.

Microsoft states that provisioned throughput is a deployment type that lets you specify the amount of throughput required for a model deployment. Unlike standard deployments, where capacity is shared and throughput can vary, a provisioned deployment holds a fixed amount of processing capacity for the deployment.

Why the other options are incorrect:

A. global batch is for asynchronous batch processing, not dedicated real-time throughput allocation.

B. data zone standard and D. standard use shared capacity; they do not allocate a fixed, specified amount of throughput to the deployment.

The correct answer is natural language processing. In this scenario, a user communicates with a chatbot using ordinary human language, such as asking, “Will it rain today?” The chatbot must interpret the meaning and intent of the user ' s text, process the request, and formulate a natural-language response containing the relevant weather information. These are characteristic natural language processing (NLP) functions.

The current AI-901 study guide explicitly includes text analysis among the common AI workloads candidates must identify, and it also covers techniques for processing and interpreting human language. Microsoft documentation on conversational language understanding explains that conversational systems interpret user utterances to determine the user ' s intent and extract relevant information before the application performs an appropriate action.

Although the weather information itself may originate from a forecasting system, the scenario being tested centers on the chatbot interaction—understanding a natural-language question and responding conversationally. Therefore, prediction and forecasting is not the best classification of the complete interaction. Computer vision is also incorrect because no images or visual information are being analyzed.

Statement 1: A summarization model is used to transcribe spoken audio into written text. = No

Summarization creates a shorter version of existing text. Transcribing spoken audio into written text is a speech recognition / speech-to-text capability, not summarization.

Statement 2: A chat completion model can generate vector embeddings for semantic tasks. = No

Chat completion models are used to generate conversational text responses. Vector embeddings for semantic search, similarity, and retrieval tasks are generated by embedding models, not chat completion models.

Statement 3: An embedding model can generate vector representations of the provided input. = Yes

This is correct. Embedding models convert input such as text into vector representations that capture semantic meaning and can be used for tasks such as similarity search, clustering, and retrieval.

Azure Content Understanding in Foundry Tools uses transcription to convert speech to text from audio and video.

Transcription is the process of converting spoken audio into written text. In Azure Content Understanding, audio and video analyzers can use transcription to produce text transcripts from spoken content.

The other options are incorrect:

diarization identifies or separates speakers, such as “Speaker 1” and “Speaker 2.”

key phrase extraction identifies important phrases from text after text exists.

synthesis converts text into spoken audio, which is the opposite direction of speech-to-text.

To call a deployed Azure OpenAI model from an application, the app must know the service endpoint and authenticate its request. Microsoft documentation states that Azure OpenAI supports API key authentication or Microsoft Entra ID authentication, and API key authentication requires including the API key in the request. Microsoft quickstart guidance also states that to successfully make a call against Azure OpenAI, you need an endpoint and a key.

The application does not need the model training dataset, the Foundry project display name, or exported playground session history to call the deployed model.

The Microsoft responsible AI principle of transparency means AI systems should be understandable. Users should know when they are interacting with AI and should be able to understand the system’s purpose, capabilities, limitations, and the basis for its decisions.

Therefore, helping users understand the decisions made by an AI system is an example of transparency.

Azure Text Analytics client library

The completed sentence is:

To develop an application that analyzes text by using Azure Language in Foundry Tools, use the Azure Text Analytics client library package.

Azure Language text analysis features include sentiment analysis, key phrase extraction, named entity recognition, language detection, and other text analytics capabilities. These are accessed by using the Azure Text Analytics client library.

The highlighted Azure Speech SDK option is incorrect because it is used for speech workloads, such as speech recognition and speech synthesis, not text analysis.

GPT and DALL-E are examples of generative AI model types/families.

GPT models generate natural language text, such as answers, summaries, emails, and conversations.

DALL-E models generate images from text descriptions.

Why the other options are incorrect:

A. transformer is a model architecture, not the best answer here as a generative model type/family.

B. computer vision is an AI workload area.

E. natural language processing is an AI workload area, not a specific generative AI model type.

The completed code is:

poller = client.begin_analyze(

analyzer_id= " invoice " ,

input_url=url

)

result = poller.result()

Azure Content Understanding analysis uses a long-running operation pattern. The Python SDK returns a poller from begin_analyze(), and Microsoft documentation states that the SDK poller handles polling automatically when you call .result().

Therefore, to retrieve the analysis results after processing completes, the correct option is:

result

The other options are incorrect because status checks operation state, wait waits without returning the final analysis object, and get_results is not the method shown for retrieving the begin_analyze() result in this code pattern.

Statement 1: Fairness can be achieved by focusing solely on improving the overall accuracy of an AI model. = No

Overall accuracy alone does not guarantee fairness. A model can have high overall accuracy while still producing biased or lower-quality outcomes for specific demographic groups. Fairness requires evaluating and reducing differences in model behavior across groups.

Statement 2: Evaluating AI system outputs to identify and reduce bias across demographic groups supports fairness. = Yes

This statement is correct. The Microsoft responsible AI principle of fairness means AI systems should treat people fairly and avoid unfair bias. Evaluating outputs across demographic groups helps identify and reduce bias.

Statement 3: Ensuring fairness for an AI system means that all users always receive the same output from the system. = No

Fairness does not mean identical outputs for all users. It means the system should avoid unfair bias and provide equitable treatment. Different users may receive different outputs when differences are relevant and appropriate.

The requirement is to produce new product images based on written descriptions. This is an image generation workload, because the AI system is creating entirely new images from natural language prompts.

Why the other options are incorrect:

B. image analysis is used to examine and interpret existing images.

C. object detection is used to identify and locate objects within an existing image.

D. optical character recognition (OCR) is used to extract text from images or scanned documents.

Since the solution must generate new visual content from user-provided descriptions, the correct answer is:

A. image generation

For a local JPEG image in a Python application using the Azure OpenAI Responses API, the image should be read, base64 encoded, and sent as a data URL, such as:

{ " type " : " input_image " , " image_url " : " data:image/jpeg;base64, < base64_image_data > " }

Microsoft’s Responses API documentation shows vision-enabled requests using input_image with image_url set to a base64 data URL in the format data:image/jpeg;base64,{base64_image}. It also confirms that JPEG images are supported.

A local file path such as file:///C:/images/photo.jpg or C:\images\photo.jpg is not a valid API-accessible image input. A public or accessible HTTPS URL can be used, but the question specifically says the application sends a local JPEG image, so the correct choice is the base64 data URL.

Azure Content Understanding analyzers define the type of content to process, such as documents, images, audio, or video, and what to extract, such as text, layout, fields, transcripts, and structured JSON output. Microsoft lists base analyzer types such as prebuilt-document, prebuilt-audio, prebuilt-video, and prebuilt-image.

For scanned invoices in PDF format, the correct choice is document analyzer, because invoices are document/form content. Microsoft’s Content Understanding quickstart specifically uses the prebuilt-invoice analyzer to extract structured data from an invoice PDF.

For voicemail recordings in WAV format, the correct choice is audio analyzer, because voicemail is audio content. Content Understanding supports structured data extraction from audio files, and the analyzer reference identifies prebuilt-audio as the base analyzer for audio content.

Statement

Answer

Defining an AI system for a typical user profile is sufficient to support inclusiveness.

No

A high number of active users means an AI system provides an inclusive experience.

No

Providing accessible interfaces and language options helps support inclusiveness in an AI system.

Yes

Comprehensive and Detailed 150 to 250 words of Explanation From Azure AI Fundamentals/Course Guide/topics:

The correct sequence is No, No, Yes.

Microsoft defines inclusiveness as designing AI systems so they work effectively for the full range of people they are intended to serve. The current AI-901 study guide explicitly includes the ability to describe considerations for inclusiveness in an AI solution as part of the Responsible AI objective domain.

The first statement is No because designing only for a hypothetical or “typical” user can exclude people with different abilities, languages, interaction styles, cultural backgrounds, or accessibility requirements. Inclusive design deliberately considers diverse and potentially underserved users.

The second statement is No because the number of active users is an adoption or usage metric. A system may have millions of users while still creating accessibility barriers or excluding particular groups. Usage volume therefore does not demonstrate inclusiveness.

The third statement is Yes. Accessible interfaces, assistive-technology compatibility, and suitable language options reduce barriers and broaden the range of people who can effectively interact with an AI system. Microsoft specifically emphasizes accessibility and supporting varied user needs as core elements of inclusive AI design.

Generating marketing email drafts from a short product description and target audience is a content generation task. This is a generative AI workload because the system creates new text based on the user’s prompt.

B. computer vision is for interpreting images or video.

C. text classification categorizes existing text, but does not draft new marketing emails.

D. speech recognition converts spoken audio into text.

Therefore, the correct answer is A. generative AI.

A recycling machine that uses cameras to identify bottles of the correct shape and reject other items requires analysis of visual input. This is a computer vision workload.

Computer vision is used for image and video analysis tasks such as object detection, classification, shape recognition, and visual inspection.

The other options are incorrect:

A. knowledge mining extracts insights from large collections of documents or data.

B. natural language processing analyzes or generates language.

C. generative AI creates new content, but the requirement is to inspect camera images and identify objects.