AI-901 Questions and Answers

Azure Content Understanding analysis operations are asynchronous long-running operations. Microsoft’s Python SDK documentation states that analysis starts with a begin operation, then the application polls for results until analysis completes.

Therefore, after submitting the PDF for analysis, the application should poll the operation results until the asynchronous analysis job succeeds.

The Microsoft responsible AI principle of transparency requires that AI systems be explainable to users.

Transparency means users should understand when they are interacting with AI, what the system can and cannot do, and how outputs or predictions are generated at an appropriate level.

The other options map to different responsible AI principles:

protect sensitive user data = Privacy and security

reduce bias in decisions = Fairness

require human oversight = Accountability

A multimodal model can receive a single user prompt that includes the image plus instructions. The prompt can ask for both outputs in one request, for example: “Analyze this image and provide a written summary and an audio summary.” Therefore, the minimum number of user prompts is 1.

Requiring human reviewers to oversee AI-generated hiring recommendations and remain responsible for the final hiring decisions is an example of the Microsoft responsible AI principle of accountability.

Accountability means people and organizations remain responsible for AI systems, their decisions, and their effects. In this scenario, the AI system assists with recommendations, but humans retain oversight and responsibility for the final decision.

The correct answer is Creating captions for a video recording. Speech recognition, also called speech-to-text, converts spoken audio into textual output. Microsoft explicitly defines Azure Speech-to-Text as speech recognition technology that transcribes audio streams or prerecorded audio into text.

Caption generation is a direct application of this capability. Microsoft describes captioning as converting the audio content of a video, film, webcast, or other production into text and displaying that text visually. Therefore, creating captions from the spoken content of a video is a canonical speech-recognition workload.

Creating an audio commentary is instead associated with speech synthesis or text-to-speech because the required output is spoken audio. Identifying key phrases in a video transcript is a text-analysis/NLP task performed after the speech has already been transcribed. A voice- activated security system may involve voice authentication, speaker recognition, or command recognition, but it is not as unambiguously a speech-to-text workload as caption generation.

The AI-901 Study Guide specifically requires candidates to distinguish the features and capabilities of speech recognition and speech synthesis.

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.

To create an AI agent that follows a specific role and behavior, you configure system instructions. Microsoft Foundry Agent Service documentation states that agent instructions define goals, constraints, and behavior.

Option A. tokens per minute (TPM) controls throughput quota, not behavior.

Option C. temperature controls response randomness/creativity, not the agent’s role.

Option D. max completion tokens controls response length, not the agent’s role or behavioral rules.

Therefore, the correct answer is B. system instructions.

Fairness

Privacy and security

Transparency

Reliability and safety

Comprehensive and Detailed Explanation with all Azure AI documents: =

Evaluating model outputs to ensure that decisions are NOT biased against specific demographic groups: Fairness

Microsoft’s responsible AI principle of fairness means AI systems should treat people fairly and avoid unfair bias across demographic groups.

Encrypting sensitive customer data and restricting system access to authorized personnel: Privacy and security

The privacy and security principle requires AI systems to protect personal and sensitive data and prevent unauthorized access.

Informing users when they are interacting with an AI system and explaining the system’s capabilities and limitations: Transparency

The transparency principle requires that people understand when they are interacting with AI and understand the system’s purpose, capabilities, and limitations.

Testing AI systems under different conditions to reduce unexpected failures: Reliability and safety

The reliability and safety principle means AI systems should operate reliably and safely, including through testing and validation under expected and unexpected conditions.

The correct parameter is analyzer_id. The Azure AI Content Understanding Python SDK uses the ContentUnderstandingClient.begin_analyze() method to submit content for processing. Microsoft defines the method signature with analyzer_id as the required identifier that specifies which analyzer should process the supplied input.

Therefore, the code should be completed as:

poller = client.begin_analyze(analyzer_id= " invoiceAnalyzer " , ...)

A custom analyzer encapsulates the configuration that determines how Content Understanding interprets the source content and which fields or structured information it extracts. Because the analyzer has already been created with the name invoiceAnalyzer, that value is supplied through analyzer_id when the analysis operation begins.

extraction_type is not the parameter used to select an analyzer. model_name would refer conceptually to an underlying AI model, but Content Understanding analysis is invoked through an analyzer abstraction rather than by specifying a model directly. Likewise, schema_id does not identify the analyzer to execute.

Microsoft ' s custom-analyzer guidance explicitly demonstrates client.begin_analyze(analyzer_id=analyzer_id, inputs=[...]), confirming the required SDK pattern

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.

Statement 1: An Azure Content Understanding in Foundry Tools analyzer returns results in the HTML format. = No

Azure Content Understanding returns analysis results as structured output, typically JSON, not HTML. Microsoft documentation states that the Content Understanding API returns analysis results in a structured JSON format.

Statement 2: The Content Understanding REST API analyzes content synchronously and returns results in the same response. = No

Content Understanding analysis is asynchronous. The quickstart states that the response header includes an Operation-Location field, which is used to retrieve the results of the asynchronous analysis operation.

Statement 3: Azure Content Understanding in Foundry Tools can transform documents, images, audio, and video files into structured output. = Yes

Microsoft describes Azure AI Content Understanding as processing documents, images, videos, and audio, and transforming them into user-defined output formats.

Statement 1: In the Foundry playground, you can upload a local image and include text in the same message when prompting a multimodal model. = Yes

Microsoft documentation for vision-enabled chat models states that in the chat session pane, you can select the attachment button, upload an image, add a text prompt such as “Describe this image,” and then submit it.

Statement 2: When using the OpenAI Responses API and a vision-enabled model, images can be provided only as base64-encoded image data. = No

Images are not limited to base64 data. The Responses API supports image input through an image URL, and Azure OpenAI documentation also describes examples using base64 encoded image data.

Statement 3: Prompts that include images require deploying a text-only model because multimodal capabilities are handled by the application layer. = No

Image prompts require a vision-enabled multimodal model, not a text-only model. Azure OpenAI vision-capable models accept image and text inputs directly.

For a vision-enabled model, Microsoft documentation shows that multimodal prompts can include both text and image content in the same user message content array. Azure OpenAI vision models accept multimodal image-and-text input and return natural language responses.

The correct request structure is a user message with a content array containing both items, for example:

{

" role " : " user " ,

" content " : [

{ " type " : " input_text " , " text " : " What is in this image? " },

{ " type " : " input_image " , " image_url " : image_url}

]

}

This avoids sending separate requests and is therefore the best choice for quickest response time.

A and C are incorrect because they send the text and image separately.

B is incorrect because the image/question should be sent as a user message, while system messages are used for instructions and behavior.

Statement 1: Using Azure Speech in Foundry Tools for batch transcription requires live audio input from a microphone. = No

Batch transcription is used to process stored audio files, not live microphone input. Live microphone input is used for real-time speech recognition scenarios.

Statement 2: Azure Speech in Foundry Tools can analyze video frames to detect and label objects while transcribing audio. = No

Azure Speech is for speech capabilities such as speech recognition, speech synthesis, and speech translation. Detecting and labeling objects in video frames is a computer vision/video analysis capability, not an Azure Speech capability.

Statement 3: Speech translation in Azure Speech in Foundry Tools can translate spoken audio into text in another language. = Yes

Speech translation converts spoken input in one language into translated text in another language.

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.

For thousands of recorded support calls stored as audio files in Azure Storage, the correct capability is speech to text batch transcription.

Microsoft states that batch transcription is designed to transcribe a large amount of audio data in storage, including audio files in Azure Blob Storage, and that files can be processed concurrently to reduce turnaround time.

Real-time transcription is for live audio, not large stored batches. Text to speech converts text into audio. Speech translation translates speech between languages, but the requirement is to generate transcripts.

Statement 1: Human-in-the-loop practices provide accountability for AI-generated decisions. = Yes

Microsoft responsible AI guidance recommends keeping a human in the loop, maintaining human oversight, and ensuring humans have a role in decisions based on model output.

Statement 2: Deploying an AI system to a production environment eliminates the need for ongoing monitoring. = No

Microsoft guidance states that after deployment, an AI-powered product or feature requires ongoing monitoring and improvement.

Statement 3: Disclosing the team that designed and deployed an AI system provides accountability for the system’s output. = Yes

Accountability in responsible AI means people and organizations remain responsible for AI systems and their effects. Identifying the people or team responsible for designing and deploying the system supports accountability and governance.

The completed sentence is:

To define an agent ' s role and behaviors, you must configure a system prompt for the agent.

In Microsoft Foundry / Azure AI agent scenarios, the system prompt, also referred to as system instructions, defines how the agent should behave, what role it should follow, and what constraints it must observe.

The other options are incorrect:

deployment slot relates to hosting/deployment routing, not agent behavior.

embedding index is used for vector search or retrieval scenarios, not defining the agent’s role.

fine tuning job changes or adapts model behavior through training, but it is not the standard configuration used to define an agent’s role and behavior at runtime.

Therefore, the correct answer is system prompt.

The correct Python method is:

recognizer.recognize_once()

Completed code:

import azure.cognitiveservices.speech as speechsdk

speech_config = speechsdk.SpeechConfig(subscription=key, region=region)

recognizer = speechsdk.SpeechRecognizer(speech_config=speech_config)

recognizer.recognize_once()

Microsoft’s Azure Speech documentation for Foundry Tools explains that Speech to text is used for real-time speech recognition and converting spoken audio into text. The Python Speech SDK uses a SpeechRecognizer object for recognition from audio input, such as a microphone.

Why the other options are incorrect:

recognizer.speak_text_async( " Ready " ) is incorrect because speaking text is text-to-speech, not speech-to-text recognition.

recognizer.start_continuous_recognition() can be used for continuous recognition, but the code shown is asking for the basic method to recognize spoken input and convert it to text from the SpeechRecognizer.

recognizer.start_keyword_recognition() is used for keyword/wake-word recognition, not general speech-to-text transcription of spoken commands.

Therefore, the correct answer is:

recognizer.recognize_once()

The output identifies text values and labels them with entity categories, such as:

now [DateTime]

students [PersonType]

Microsoft [Organization]

175 [Quantity-Number]

Today [DateTime-Date]

This is entity recognition, also known as Named Entity Recognition (NER). NER identifies and categorizes entities in unstructured text, such as people, organizations, locations, dates, quantities, numbers, and other named or structured values.

Why the other options are incorrect:

B. translation would convert text from one language to another.

C. key phrase extraction would return important phrases or main topics, not categorized entity labels.

D. sentiment analysis would return sentiment such as positive, negative, neutral, or mixed.

Statement 1: Evaluators in Microsoft Foundry replace the need for configuring token limits. = No

Evaluators are used to assess model or agent output quality, safety, and reliability. They do not replace model configuration settings such as token limits. Microsoft describes evaluators as tools that measure the quality, safety, and reliability of AI responses.

Statement 2: Evaluators in Microsoft Foundry can assess the quality and safety of responses generated by a generative AI model. = Yes

This is correct. Microsoft Foundry includes built-in evaluators for general quality metrics such as coherence and fluency, safety/security metrics, RAG metrics, and agent-specific metrics.

Statement 3: Evaluators in Microsoft Foundry can retrain a deployed generative AI model automatically when quality issues are detected. = No

Evaluators measure and report quality, safety, and reliability issues. They do not automatically retrain deployed generative AI models. Microsoft describes evaluation as measuring model or agent performance against test data, while monitoring can alert when outputs fail quality thresholds.

No

Yes

No

Comprehensive and Detailed Explanation with all Azure AI documents: =

Statement 1: Speech synthesis converts spoken audio into written text. = No

Speech synthesis, also called text to speech, converts written text into spoken audio. Converting spoken audio into written text is speech recognition or speech to text.

Statement 2: Speech recognition can detect and transcribe spoken input. = Yes

Speech recognition is used to identify spoken language in audio and convert it into text.

Statement 3: Text to speech in Azure Speech in Foundry Tools can translate text input into audible speech in a different language. = No

Text to speech converts text into spoken audio. Translation into another language is a separate translation capability, not standard text-to-speech synthesis.

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.

Azure Vision in Foundry Tools supports OCR capabilities for reading printed and handwritten text from images and documents. Microsoft documentation states that OCR extracts text from images and documents, including handwritten text.

Therefore, extracting data from handwritten letters is the action that can be performed by using Azure Vision in Foundry Tools.

Statement 1: Generating a response to a user prompt occurs during the inference stage. = Yes

When a deployed generative AI model receives a user prompt and generates an output, that process is called inference. The model is being used to produce a response, not being trained.

Statement 2: A generative AI model generates responses by copying stored documents directly from the model’s training data. = No

Generative AI models do not normally respond by simply copying stored documents from training data. They generate outputs based on learned statistical patterns. In some cases, models may reproduce memorized content, but that is not the normal or intended generation mechanism.

Statement 3: A generative AI model produces output by predicting the next token based on patterns learned from the model’s training data. = Yes

This is correct. Large language models generate text by predicting likely next tokens based on the prompt and learned patterns from training.

The Model catalog is used for comparing and deploying a wide range of models for generative AI development in Microsoft Foundry.

Microsoft documentation states that the model catalog in Foundry portal is the hub for discovering and using a wide range of models to build generative AI applications. It also includes many models across providers such as Azure OpenAI, Mistral, Meta, Cohere, NVIDIA, and Hugging Face.

The Microsoft Learn module for Foundry models also states that the model catalog is used to explore and filter models, compare models using benchmark metrics, and deploy a model to an endpoint.

To minimize cost and keep generated responses within a defined length, configure Max Completion Tokens.

Microsoft’s Azure OpenAI / Foundry API reference defines max_completion_tokens as an upper bound for the number of tokens that can be generated for a completion. Because generated tokens contribute to usage and response length, limiting completion tokens helps control both output length and cost.

Temperature and Top P control randomness or sampling behavior, not maximum response length. Model version settings do not directly define the generated response length.

The correct answer is API endpoint. After a generative AI model is deployed in Microsoft Foundry, the deployment exposes an inference endpoint through which external applications can submit requests and receive generated responses. Microsoft Foundry documentation describes deployed models as being accessible programmatically through an endpoint and associated credentials. Applications invoke the relevant REST API or SDK against this endpoint and specify the deployed model as required by the API.

An embedding vector is a numerical representation of semantic information used for similarity search, retrieval, and grounding; it is not the network interface through which a deployed model is invoked. A training dataset contains data used to train or fine-tune a model and is not called when performing inference. A URL parameter can be part of an API request, but it does not represent the deployed model ' s callable interface.

This directly aligns with the AI-901 objective to deploy a model and interact with it in the Foundry portal and to create lightweight applications that invoke deployed generative AI models through supported APIs and SDKs.

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 " )

Statement 1: The Temperature parameter can be set before deploying a model. = No

temperature is an inference/request parameter used when calling or testing a deployed model. It controls randomness in generated responses. It is not a required setting for deploying the model itself.

Statement 2: During inference, the model name is used to route requests to a specific deployment. = No

In Azure OpenAI / Microsoft Foundry deployments, application requests are routed to a specific deployment name, even when the SDK parameter is called model. The underlying model name, such as gpt-4.1-mini, is not what routes the request to the deployment.

Statement 3: After a model is deployed, both code and testing tools can be used to interact with the model. = Yes

After deployment, you can test the model in Foundry playground/testing tools or call the deployment from application code by using the endpoint, deployment name, and authentication credentials.

input=[

{

" role " : " user " ,

" content " : [

{ " type " : " input_text " , " text " : " What is in this image? Provide 3 bullet points. " },

{ " type " : " input_image " , " image_url " : image_url},

],

}

]

print(response.output_text)

For a multimodal request that sends both text and an image to an Azure OpenAI model, the input content array uses separate content item types. Text content is sent with type: " input_text " , and image content is sent with type: " input_image " . Microsoft’s Azure OpenAI Responses API documentation shows the Responses API supports multimodal inputs and uses structured input content items in the request.

The options output_text and output_image are not used inside the user input content array. In the shown code, response.output_text is used after the response is returned to print the model’s text answer.