Practice Mode

AI-901

Loading...

AI-901 — 20 free practice questions

Preview the exam for free. Unlock all 52 questions →

Question 1

You are using the Azure Speech SDK to develop a Python application that supports real-time spoken conversations. Which Azure Speech class should you use to configure the connection to the Azure Speech service? Options: A. AudioOutputConfig B. SpeechSynthesizer C. AudioConfig D. SpeechConfig
  • A. AudioOutputConfig
  • B. SpeechSynthesizer
  • C. AudioConfig
  • D. SpeechConfig Correct

Explanation

SpeechConfig is the class that encapsulates the connection settings to the Azure Speech service, holding the subscription key/token and the region (or endpoint). It must be created first and passed to recognizers and synthesizers so they know which resource and endpoint to use. AudioConfig only specifies the audio input/output source (microphone, speaker, or file), SpeechSynthesizer performs text-to-speech rather than configuring the connection, and AudioOutputConfig relates to routing synthesized audio output.

Question 2

Select the answer that correctly completes the sentence.

Explanation

With the OpenAI Responses API and a vision-enabled model, an image can be passed inline by embedding it directly in the request as base64-encoded image data within the message content. This avoids hosting the file externally, since the encoded bytes are sent as part of the input payload. The alternative of a publicly reachable image URL is also supported, but the base64 data option lets you include local images directly in the request.

Question 3

You need to create an AI agent in Microsoft Foundry that follows a specific role and behavior when responding to users. What should you configure? Options: A. system instructions B. temperature C. tokens per minute (TPM) D. max completion tokens
  • A. system instructions Correct
  • B. temperature
  • C. tokens per minute (TPM)
  • D. max completion tokens

Explanation

System instructions define the persona, role, tone, and behavioral rules an agent follows when responding, making them the correct place to enforce a specific role and behavior. Temperature only controls randomness/creativity of outputs, tokens per minute (TPM) is a throughput quota, and max completion tokens caps the length of a response, none of which shape the agent's role or conduct.

Question 4

For each of the following statements, select Yes if the statement is true. Otherwise, select No. NOTE: Each correct selection is worth one point.

Explanation

Statement 1 is No because system prompts guide model behavior and cannot enforce identity or access control; authorization must be handled by authentication and authorization mechanisms outside the prompt. Statement 2 is No because a system prompt does not manage or reduce the TPM quota, which is a service-level throughput setting. Statement 3 is Yes because the core purpose of a system prompt is to set context, role, and constraints that guide how the generative AI model responds.

Question 5

You are developing a web app that processes invoices to calculate expenses. You need to extract structured fields, including nested values, from the invoices by using a defined schema. What should you use? Options: A. an optical character recognition (OCR)-only document processing pipeline B. a transcription workflow in Azure Speech in Foundry Tools C. an analyzer in Azure Content Understanding in Foundry Tools D. an Azure AI Search service
  • A. an optical character recognition (OCR)-only document processing pipeline
  • B. a transcription workflow in Azure Speech in Foundry Tools
  • C. an analyzer in Azure Content Understanding in Foundry Tools Correct
  • D. an Azure AI Search service

Explanation

An analyzer in Azure Content Understanding lets you define a schema and extract structured fields, including nested and hierarchical values, from documents such as invoices. An OCR-only pipeline merely returns raw text with no schema-based structured extraction, Azure Speech handles audio transcription rather than documents, and Azure AI Search provides indexing and retrieval rather than schema-driven field extraction.

Question 6

For each of the following statements, select Yes if the statement is true. Otherwise, select No. NOTE: Each correct selection is worth one point.

Explanation

Statement 1 is No because Voice Live is a full conversational solution that produces spoken responses, not just transcribed text. Statement 2 is No because Voice Live integrates the speech-to-text and text-to-speech capabilities for you, removing the need to wire them up separately. Statement 3 is Yes because Voice Live unifies speech to text, model reasoning, and text to speech into a single low-latency conversational experience.

Question 7

You are developing a voice application that listens for spoken commands and converts them into text by using Azure Speech in Foundry Tools. How should you complete the Python code? To answer, select the appropriate option in the answer area. NOTE: Each correct selection is worth one point.

Explanation

After creating a SpeechConfig and a SpeechRecognizer, calling recognizer.recognize_once() performs a single synchronous recognition, capturing one spoken utterance and returning it as text, which fits an application that listens for individual spoken commands. This method waits for a single result rather than continuous streaming, making it the appropriate call to complete the recognition step.

Question 8

What are two purposes of instructions when prompting a generative AI model? Each correct answer presents part of the solution. NOTE: Each correct selection is worth one point. Options: A. defines the agent’s role and behavior B. selects which model to use C. defines the Azure region where inference occurs D. defines the tokens per minute (TPM) allocation for the model E. defines constraints on the model’s responses
  • A. defines the agent’s role and behavior Correct
  • B. selects which model to use
  • C. defines the Azure region where inference occurs
  • D. defines the tokens per minute (TPM) allocation for the model
  • E. defines constraints on the model’s responses Correct

Explanation

Instructions in a prompt define the agent's role, persona, and behavior (A) and set constraints or guardrails on the responses the model may produce (E), shaping both how it acts and what it may output. Selecting the model, choosing the Azure region for inference, and setting the TPM allocation are deployment and configuration decisions made outside the prompt, so B, C, and D are not purposes of prompt instructions.

Question 9

You are developing an application that converts text into spoken audio and saves the synthesized audio to a file by using Azure Speech in Foundry Tools. How should you complete the Python code? To answer, select the appropriate option in the answer area. NOTE: Each correct selection is worth one point.

Explanation

To save synthesized speech to a file, the audio output must be directed to a file rather than a speaker. AudioOutputConfig(filename="output.wav") creates an audio configuration that writes the generated audio to the specified WAV file, which is then passed to the SpeechSynthesizer. Using a device or default speaker configuration would play the audio instead of persisting it.

Question 10

Select the answer that correctly completes the sentence.

Explanation

An agent's role, personality, and behavioral boundaries are defined through its system prompt (also called system instructions). This instruction is sent to the underlying generative model to shape how it responds and what actions it takes throughout the conversation. User prompts drive individual requests, but the system prompt establishes the persistent identity and rules for the agent.

Question 11

You are developing an application that analyze invoices by using Azure Content Understanding in Foundry Tools. You need to ensure that the application retrieves the analysis results after processing completes. How should you complete the Python code? To answer, select the appropriate option in the answer area. NOTE: Each correct selection is worth one point.

Explanation

With the long-running (poller) pattern in the Azure SDK, begin_analyze returns a poller object, and calling poller.result() blocks until the operation completes and then returns the final analysis results. This is exactly what is needed to retrieve results after processing finishes. Methods like status() or done() only report progress rather than returning the completed output.

Question 12

You have a Microsoft Foundry project named project1 that contains an Azure OpenAI resource named Resource1. To Resource1, you deploy a gpt-4.1-mini model by using a model deployment named my-mini-gpt. You need to connect to my-mini-gpt from an application. How should you complete the Python code? To answer, drag the appropriate values to the correct targets. Each value may be used once, more than once, or not at all. NOTE: Each correct selection is worth one point. | Answer Area | | |---|---| | client = OpenAI(api_key="...", | | | base_url="https:// | .openai.azure.com/openai/v1/ | | response = client.responses.create( | | | model=" | “, | | 1 | |

Explanation

The base_url follows the pattern https://<resource-name>.openai.azure.com/openai/v1/, so the Azure OpenAI resource name resource1 belongs in the host portion of the endpoint. The model parameter must reference the deployment name rather than the base model name, so my-mini-gpt (the deployment) is used, not gpt-4.1-mini. Azure OpenAI routes requests by deployment name, which is why the deployment identifier is required here.

Question 13

For each of the following statements, select Yes if the statement is true. Otherwise, select No. NOTE: Each correct selection is worth one point.

Explanation

"An AI generative model is retrained before each request" is No, because models are trained once and then reused for inference; retraining per request would be prohibitively expensive and is not how they work. "An AI agent copies and pastes answers from a database" is No, because agents use generative models to produce responses dynamically rather than returning fixed stored records. "An AI agent uses a generative AI model to establish actions based on user input" is Yes, since agents rely on a generative model to interpret input and decide which actions or tools to invoke.

Question 14

You are reviewing best practices for using AI at your company. Which Microsoft responsible AI principle is each task an example of? To answer, drag the appropriate principles to the correct tasks. Each task may be used once, more than once, or not at all. You may need to drag the split bar between panes or scroll to view content. NOTE: Each correct match is worth one point. AnswerArea Evaluating model outputs to ensure that decisions are NoT biased against specific demographic groups: Encrypting sensitive customer data and restricting system access to authorized personnel: Informing users when they are interacting with an Al system and explaining the system's capabilities and limitations: Testing Al systems under different conditions to reduce unexpected failures:

Explanation

Evaluating outputs to avoid bias against demographic groups maps to Fairness, which requires AI systems to treat all people equitably. Encrypting sensitive data and restricting access maps to Privacy and security, which protects data and safeguards systems from misuse. Informing users they are interacting with AI and explaining capabilities and limitations maps to Transparency, ensuring systems are understandable. Testing under varied conditions to reduce unexpected failures maps to Reliability and safety, which ensures systems perform dependably even in unforeseen situations.

Question 15

For each of the following statements, select Yes if the statement is true. Otherwise, select No. NOTE: Each correct selection is worth one point.

Explanation

"Generating a response to a user prompt occurs during inference" is Yes, because inference is the stage where a trained model is used to produce output. "A model generates responses by copying stored documents from training data" is No, since generative models learn statistical patterns rather than storing and retrieving verbatim documents. "A model produces output by predicting the next token based on learned patterns" is Yes, which describes exactly how autoregressive generative language models work.

Question 16

You need to build an AI solution that produces new product images based on written descriptions provided by users. Which AI workload should you use? Options: A. image analysis B. image generation C. object detection D. optical character recognition (OCR)
  • A. image analysis
  • B. image generation Correct
  • C. object detection
  • D. optical character recognition (OCR)

Explanation

Producing entirely new images from written text descriptions is the definition of image generation, a generative AI workload where a model synthesizes novel visuals from a prompt. Image analysis, object detection, and OCR are all interpretive workloads that examine existing images to extract information such as tags, bounding boxes, or text, and none of them create new images.

Question 17

Select the answer that correctly completes the sentence.

Explanation

Detecting and reading text within scanned documents and images is a computer vision capability, specifically optical character recognition (OCR). Computer vision analyzes visual pixel data to locate and extract printed or handwritten text, which underpins information extraction and document processing solutions. Other AI workloads such as natural language processing operate on already-digitized text rather than interpreting the images themselves.

Question 18

You have a Microsoft Foundry project that has a generative AI model deployment. You need to ensure that responses generated by the model minimize costs and remain within a defined length. Which parameter should you configure? Options: A. Top P B. Temperature C. Max Completion Tokens D. Model version settings
  • A. Top P
  • B. Temperature
  • C. Max Completion Tokens Correct
  • D. Model version settings

Explanation

Max Completion Tokens caps the number of tokens a model can generate in its response, directly limiting output length and controlling cost, since billing is based on tokens produced. Top P and Temperature only influence the randomness and diversity of word selection, not how long the response is. Model version settings determine which build of the model runs and do not constrain response length.

Question 19

For each of the following statements, select Yes if the statement is true. Otherwise, select No. NOTE: Each correct selection is worth one point.

Explanation

Human-in-the-loop practices are Yes because keeping people responsible for reviewing and approving AI decisions establishes clear accountability. Deploying to production eliminating the need for monitoring is No, as responsible AI requires ongoing monitoring to catch drift, errors, and unexpected behavior after release. Disclosing the team that designed and deployed the system is Yes, since transparency about who is responsible reinforces accountability for its outputs.

Question 20

Your company processes customer support emails. You need to implement an AI solution that automatically identifies mentions of people, organizations, and locations in the emails. Which text analysis technique should you use? Options: A. Named Entity Recognition (NER) B. key phrase extraction C. sentiment analysis D. summarization
  • A. Named Entity Recognition (NER) Correct
  • B. key phrase extraction
  • C. sentiment analysis
  • D. summarization

Explanation

Named Entity Recognition (NER) identifies and classifies specific real-world entities in text, such as people, organizations, and locations, which is exactly the requirement. Key phrase extraction only surfaces main talking points without categorizing them by type, sentiment analysis gauges positive or negative tone, and summarization condenses content. Only NER labels the mentioned entities by category.

Verifying access...