AI-901 Practice Questions

A vision-enabled endpoint can receive an image either (1) as a public URL that the service can download itself, or (2) as the raw picture data embedded in the request.

For online endpoints that accept JSON payloads, the raw picture is normally supplied as a base-64-encoded string; for REST calls that accept an octet-stream body, the client sends the same bytes but they are often first converted to base-64 before being placed in the request payload.

Therefore the two supported request formats are a publicly reachable URL (C) and a base-64-encoded image string (D).

A. A JSON document that merely “describes” the content supplies no actual image bytes; the model cannot analyze non-image descriptive text.

B. UTF-8 text captions are metadata, not pixel data; vision models require the image itself, not a textual description.

1. Microsoft, “Quickstart: Deploy and score a vision model with Azure Machine Learning online endpoints”, section ‘Prepare the request – encode the image as base64’, docs.microsoft.com (2023-10-31).

2. Microsoft Cognitive Services, “Analyze Image (Computer Vision REST API v3.2) – Request parameters”, notes ‘Image URL’ OR ‘binary image data’, docs.microsoft.com (2023-09-14).

3. MIT Professional Education & Azure, “AI on Azure: Vision-enabled services”, lecture slides 5-6, showing URL or base64 payload options (Spring 2023).

In Microsoft-hosted OpenAI experiences such as Foundry, an agent’s role, tone, and permissible actions are defined inside the chat request’s system-level prompt (often called a “system instruction” or “system message”). The model reads this instruction first and uses it as a governing rule set for every subsequent user interaction, ensuring the agent consistently follows the required role and behavioral constraints. Settings such as temperature, token limits, or throughput affect style or resource usage but do not, by themselves, impose a role or behavioral policy.

A. tokens per minute (TPM) – Controls request-rate throttling; has no impact on the semantic behavior of a single agent.

C. temperature – Adjusts randomness/creativity of outputs; does not specify role, persona, or rules.

D. max completion tokens – Sets length cap on a reply; does not influence the role or behavioral guidelines.

1. Microsoft Learn – “Chat completions: system messages define the assistant’s behavior.” Section “System messages,” para. 1 (https://learn.microsoft.com/azure/ai-services/openai/how-to/chat)

2. Microsoft Learn – “Prompt engineering techniques…Use a system message to set the overall instructions for the assistant.” Section “Give the model context,” bullets 1-2 (https://learn.microsoft.com/azure/ai-services/openai/concepts/prompt-engineering)

3. Microsoft Copilot Foundry documentation – “Create an agent…Add a system instruction to establish its role and governing rules.” Guide v1.0, chap. 2, p. 5.

4. Stanford CS25 Lecture Notes – “System prompt anchors the assistant’s persona and constraints” (Spring 2024, Lec 04, slide 14).

In the Azure Speech SDK (Python), you establish and configure the connection to the Azure Speech service with the SpeechConfig class (e.g., SpeechConfig.fromsubscription or fromendpoint). AudioConfig only specifies the audio source/ sink, SpeechSynthesizer performs text-to-speech once a SpeechConfig is created, and AuditOutputConfig controls audit‐log output. None of the listed options configures the network connection or authentication parameters; therefore no listed choice is correct.

A. AudioConfig – manages microphone, stream, or file I/O; contains no keys, region, or endpoint settings.

B. SpeechSynthesizer – consumes an existing SpeechConfig to synthesize speech; does not hold connection settings itself.

C. AuditOutputConfig – directs where audit logs are stored; unrelated to service connection or authentication.

1. Microsoft Docs – “Create a SpeechConfig object” (azure.cognitiveservices.speech) Section “Configure the Speech SDK”, para 2. https://learn.microsoft.com/azure/cognitive-services/speech-service/how-to-configure-sdk

2. Microsoft Docs – Quickstart: “Recognize speech from a microphone with Python”, code sample lines 16-19 (shows SpeechConfig usage). https://learn.microsoft.com/azure/cognitive-services/speech-service/quickstart-python?pivots=programming-language-python#recognize-speech-from-a-microphone

3. Microsoft Docs – AudioConfig class reference, states purpose is audio source/sink only (Class summary, first paragraph). https://learn.microsoft.com/python/api/azure-cognitiveservices-speech/azure.cognitiveservices.speech.audioconfig

4. Microsoft Docs – SpeechSynthesizer class reference, “Requires a SpeechConfig” (Class description, first paragraph). https://learn.microsoft.com/python/api/azure-cognitiveservices-speech/azure.cognitiveservices.speech.speechsynthesizer

5. Microsoft Docs – AuditOutputConfig class reference, describes audit logging only (Overview section). https://learn.microsoft.com/python/api/azure-cognitiveservices-speech/azure.cognitiveservices.speech.translation.audioconfig

For vision-enabled model deployments, the image to be analysed must be supplied as a separate content item whose type is inputimage. When the Azure OpenAI Responses (batch) API receives a message that contains a content array, it looks for an item marked inputimage and passes the referenced or in-line image to the vision model; any other type name is ignored or considered invalid for analysis. Therefore, inputimage is the only valid key for providing the image itself.

A. imagebase64 – Not a recognised content-item type; base-64 data must still be wrapped in an inputimage item.

B. imagegeneration – Reserved for DALL-E image creation tasks, not accepted when submitting an image for analysis.

C. outputimage – Used by the service when it returns generated images; it cannot be used in the request body.

1. Microsoft Learn: “Call vision-enabled deployments with the Requests/Responses API”, section “content items”, bullet “inputimage” – learn.microsoft.com/azure/ai-services/openai/how-to/vision#content-items

2. Microsoft Learn REST reference: “chat/completions – request schema”, 2024-03-01-preview, Table 2 “type enumeration (text | inputimage | imageurl | …)” – learn.microsoft.com/rest/api/ai-services/openai/chat-completions

3. Microsoft Learn: “Using batch Requests/Responses with multimodal models”, Example 1 JSON (lines showing "type":"inputimage") – learn.microsoft.com/azure/ai-services/openai/how-to/batch-requests-responses#multimodal-example

https://kxbjsyuhceggsyvxdkof.supabase.co/storage/v1/object/public/file-images/Microsoft_AI-901/page_21_img_2.jpg

Statement 1: In the new Microsoft Foundry portal, you must fine-tune a model before you can deploy

the model. = No

Fine-tuning is optional. Microsoft’s Foundry model deployment documentation describes deploying

Foundry Models directly from the model catalog for inference. It does not require fine-tuning first.

Statement 2: In the new Microsoft Foundry portal, you can test a model from the model catalog only

after you deploy the model. = No

Microsoft documentation states that some Foundry Tools are available to try via the model catalog

without a project, and Foundry playgrounds are used for prototyping and validation before

production. Therefore, the statement using “only after you deploy” is too restrictive.

Statement 3: In the new Microsoft Foundry portal, you can deploy a model from the model catalog

only after retraining the model. = No

Retraining/fine-tuning is not required before deployment. Microsoft states that after you deploy a

Foundry Model, you can interact with it in the Foundry Playground and use it from code, and the

deployment workflow starts by selecting a model from the model catalog and choosing Deploy.

https://kxbjsyuhceggsyvxdkof.supabase.co/storage/v1/object/public/file-images/Microsoft_AI-901/page_32_img_1.jpg

The completed sentence is:

Information extraction solutions that detect and read text in scanned documents and images rely on

computer vision.

Detecting and reading text in scanned documents and images is typically done with OCR, which is a

computer vision capability. Microsoft describes OCR as text recognition/text extraction that extracts

printed or handwritten text from images and documents.

https://kxbjsyuhceggsyvxdkof.supabase.co/storage/v1/object/public/file-images/Microsoft_AI-901/page_9_img_2.jpg

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.

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.

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://.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