Q1.A development team is deploying a vision-capable model in a Microsoft Foundry project.
They are building an application that sends messages containing both text and an image URL.
The solution must keep response times as short as possible.
What message structure should be used for the request?
Show answer
With a vision-capable model, text and images can be sent together in the content array of the same user message.
Microsoft Learn also shows an example of handling image input containing an image URL and text input in a single request.
Option D, which includes both in the same content array within the user message rather than splitting them into separate requests, achieves the shortest response time.
The system message is used mainly for instructions to the model and is not suited to carrying the image question itself.
Use vision-capable chat models
Q2.You are developing a Python application that supports real-time voice conversations using the Azure Speech SDK.
Which Azure Speech class should you use to configure the connection to the Azure Speech service?
Show answer
The answer key for this question is B, but on Microsoft Learn the connection information for the Speech service is normally configured with SpeechConfig, while AudioOutputConfig is the class that configures the output destination for speech synthesis.
Among the choices, the closest is AudioOutputConfig, which handles the output configuration for speech synthesis.
AudioConfig configures the audio for input or output, and SpeechSynthesizer is the class that performs the synthesis, so neither is the primary component that configures the connection itself.
AudioOutputConfig class
Q3.You are developing an application that analyzes voicemail recordings using Azure Content Understanding in Foundry Tools.
You need to extract a transcription and structured information from the recordings.
Which kind of analyzer should you choose?
Show answer
Because a voicemail recording is audio data, the analyzer to use is the audio analyzer.
Microsoft Learn explains that the Content Understanding audio analyzer transcribes conversational speech and performs speaker separation, and can extract structured fields such as summaries, sentiment, and key topics.
The analyzers for images, video, and documents target different types of input data and are therefore not suited to this requirement.
Azure Content Understanding for Foundry Tools audio solutions
Q4.You are organizing best practices for applying AI in your company.
Each task is an example of which of Microsoft’s responsible AI principles?
To answer, drag the appropriate principle onto the correct task.
Each principle may be used once, more than once, or not at all.
Drag each option to the matching requirement’s drop zone, or tap to place it. Click a filled zone to send it back. (The same option may be used more than once.)
Fairness
Inclusiveness
Privacy and security
Reliability and safety
Transparency
Show answer
In Microsoft’s responsible AI principles, reducing bias to aim for fair decisions maps to Fairness, and data protection with access control maps to Privacy and security.
In addition, disclosing the use of AI and explaining capabilities and limitations is Transparency, and improving stable operation and safety through thorough testing is Reliability and safety.
It is important to be able to map each principle to a concrete operational example.
Microsoft’s responsible AI
Q5.A Microsoft Foundry project has a deployment of a vision-capable model.
You send prompts to the model using the Azure OpenAI Responses API.
You need to pass an image for analysis.
Which content item should you include in the request?
Show answer
When passing an image to a vision-capable model with the Azure OpenAI Responses API, the image is treated as input content, so you include input_image in the content array.
The expected structure specifies an image_url to pass the image to be analyzed.
image_generation is for generating images, output_image is an output image, and image_base64 is not the official content item for this question.
Therefore, input_image is the correct choice for including an image for analysis in the request.
Use the Azure OpenAI Responses API
Q6.You need to build an AI solution that drafts marketing emails from a short description of a product and its target audience.
Which AI workload should you use?
Show answer
Taking a short product description and target audience as input and producing new text such as a marketing email draft is a representative use of generative AI.
Generative AI creates new content such as text, images, and code based on a prompt.
Computer vision analyzes images, text classification categorizes existing text, and speech recognition transcribes audio, so none of them apply to generating new email content.
Introduction to generative AI and agents
Q7.A Microsoft Foundry project has a deployment of a generative AI model.
You need to minimize the cost of the responses the model generates while keeping them within a defined length.
Which parameter should you configure?
Show answer
To control the length of a generative AI model’s response and limit the number of tokens used in the output, you configure Max Completion Tokens.
Because the number of tokens generated affects usage and cost, setting an upper limit makes it easier to keep responses within a defined length.
Temperature and Top P are settings that adjust the randomness of the output and do not directly limit the maximum response length.
Model version settings are also not used to control response length.
Azure OpenAI REST API reference for Microsoft Foundry Models
Q8.A Microsoft Foundry project has a deployment of a generative AI model.
You use the Foundry playground to check the model’s behavior.
You need to develop an application that sends requests to the deployed model.
What information should the application include in order to call the model?
Show answer
To call a deployed Azure OpenAI model from an application, you need the model endpoint to which the request is sent, and authentication credentials such as an API key or Microsoft Entra ID.
The training dataset, the display name of the Foundry project, and the playground session history are not used for authenticating or connecting to call the model.
Note that Azure AI Studio has now been consolidated and renamed to Azure AI Foundry.
Use the Azure OpenAI Responses API
Q9.You are developing an application that processes voicemail recordings using Azure Content Understanding in Foundry Tools.
Which capability does Azure Content Understanding use to convert speech to text?
Show answer
The capability that converts audio data to text is transcription.
The audio analysis in Azure Content Understanding can convert audio content such as conversations and recordings into searchable, analyzable text transcripts.
Voice Live is for real-time voice interaction, key phrase extraction extracts important terms, and OCR recognizes characters in images or documents, so none of them apply to converting voicemail audio to text.
Azure Content Understanding for Foundry Tools audio solutions
Q10.You are building a Conversational Language Understanding (CLU) model for an e-commerce company.
You need to enable the model to detect utterances that fall outside its intended scope.
What should you do?
Show answer
Conversational Language Understanding (CLU) provides a special intent called the None intent, which is used to classify utterances that do not match any custom intent.
To correctly detect out-of-scope questions and unrelated input, it is important to add representative example utterances to the None intent and train the model on them.
This enables the model to classify user input as None when it does not match any existing intent.
None intent