Microsoft Azure AI Fundamentals

Question 1

You are building a single-agent solution in the Microsoft Foundry portal for a retail company. The agent must answer customer questions and, when asked about product availability, look up live stock levels from the company's inventory REST API. Which capability should you configure for the agent so it can retrieve this real-time data?

Accepted Answer

Add a tool to the agent that calls the inventory API

Answer

Increase the model's max tokens so it can store inventory data

Answer

Lower the temperature setting to make responses more factual

Answer

Add the current inventory list into the agent's system prompt

Question 2

A media company wants to automatically produce written descriptions of the visual content in its archive of video clips, including identifying objects, actions, and scenes so editors can search the archive by content. Which AI workload category best addresses this requirement?

Accepted Answer

Computer vision

Answer

Speech recognition

Answer

Sentiment analysis

Answer

Keyword extraction

Question 3

A company wants to build a lightweight app that captures live audio from an in-person staff meeting through a microphone and displays a running text transcript on screen as people speak. The developers plan to use Azure Speech in Foundry Tools. Which Speech capability should they use to meet this requirement?

Accepted Answer

Real-time speech-to-text transcription

Answer

Text-to-speech synthesis

Answer

Speech translation to another language

Answer

Batch transcription of stored audio files

Question 4

A developer wants to add a chat feature to an internal Python web app. A GPT model is already deployed in an Azure AI Foundry project. The developer needs the app to send user messages to the deployed model and receive responses programmatically, without building any infrastructure from scratch. Which approach best meets this requirement?

Accepted Answer

Use the Azure AI Foundry SDK to connect to the project and call the deployed model's chat completions endpoint from the Python code.

Answer

Fine-tune a new model in the Foundry portal each time a user sends a message.

Answer

Export the model weights from Foundry and host the model on a local server inside the web app.

Answer

Use Azure Content Understanding to process the chat messages and return responses.

Question 5

A developer at a logistics company has already created and tested a single-agent solution in the Azure AI Foundry portal that answers shipment-tracking questions. They now want to embed this agent's capabilities into their internal web application so employees can interact with it programmatically, without opening the Foundry portal. What is the most appropriate way to accomplish this?

Accepted Answer

Build a lightweight client application that connects to the existing agent using the Foundry SDK (Agents client)

Answer

Retrain the underlying base model with the company's shipment data to add tracking capabilities

Answer

Export the agent as a static container image and host it as a standalone website

Answer

Recreate the agent's logic manually inside the web application using only regular expressions

Question 6

A call center wants to automatically convert recorded customer support phone calls into written text so that supervisors can search and review the conversations later. Which AI workload should the company use to accomplish this?

Accepted Answer

Speech recognition

Answer

Speech synthesis

Answer

Sentiment analysis

Answer

Entity detection

Question 7

A software company receives thousands of customer support emails daily. They want to automatically route each email to the correct department by determining which of five predefined categories (Billing, Technical, Sales, Legal, or General) best describes the email content. Which AI capability should they use to accomplish this?

Accepted Answer

Text classification

Answer

Named entity recognition

Answer

Sentiment analysis

Answer

Text summarization

Question 8

A financial services company builds an AI assistant that answers customer questions about investment products. Compliance requires that whenever the assistant provides an answer, it must clearly indicate which internal documents the information came from and note that the response was generated by AI. Which responsible AI principle does this requirement primarily support?

Accepted Answer

Transparency

Answer

Reliability and safety

Answer

Privacy and security

Answer

Inclusiveness

Question 9

A city government is deploying an AI-powered chatbot to help residents access public services. The team wants to ensure the chatbot works effectively for residents with visual impairments, limited literacy, and those who speak languages other than the region's primary language. Which responsible AI principle is the team primarily addressing?

Accepted Answer

Inclusiveness

Answer

Reliability and safety

Answer

Privacy and security

Answer

Accountability

Question 10

A bank deploys a generative AI assistant to answer customer questions about account fees. During testing, the team notices that the same question sometimes produces conflicting fee amounts, which could mislead customers. Which responsible AI principle is the team primarily addressing when they work to ensure the assistant returns consistent, dependable answers?

Accepted Answer

Reliability and safety

Answer

Inclusiveness

Answer

Privacy and security

Answer

Transparency

Question 11

A news organization wants to automatically generate a short paragraph that captures the main points of each lengthy article so readers can quickly decide whether to read the full text. Which text-analysis technique should they use?

Accepted Answer

Summarization

Answer

Keyword extraction

Answer

Entity detection

Answer

Sentiment analysis

Question 12

A development team is building an internal knowledge search tool. They want user queries and documents to be converted into numeric vector representations so that documents with similar meaning to a query can be retrieved, even when they do not share the exact same words. Which type of model should they use to generate these representations?

Accepted Answer

An embedding model

Answer

An image-generation model

Answer

A speech-recognition model

Answer

An object-detection model

Question 13

A software development team wants to integrate an AI capability into their internal developer portal that generates working code snippets and functions from plain-language descriptions typed by engineers. Which type of generative AI model is best suited for this requirement?

Accepted Answer

A large language model capable of code generation

Answer

An image-generation model that produces visuals from text prompts

Answer

A speech-recognition model that transcribes audio to text

Answer

A sentiment-analysis model that classifies text tone

Question 14

A marketing analyst wants to automatically group thousands of customers into segments based on similarities in their purchasing behavior, without providing any predefined labels or categories. Which type of machine learning workload best fits this requirement?

Accepted Answer

Clustering

Answer

Classification

Answer

Regression

Answer

Sentiment analysis

Question 15

You are building a single-agent solution in the Microsoft Foundry portal for a finance team. Users will paste raw quarterly sales numbers into the chat and ask the agent to calculate growth percentages, produce a bar chart, and return the results. The agent's language model cannot reliably perform arithmetic or generate images on its own. Which capability should you enable for the agent so it can run calculations and create the chart?

Accepted Answer

The Code Interpreter tool

Answer

A higher temperature setting on the model

Answer

The File Search (knowledge) tool grounded on company documents

Answer

A larger maximum token limit for responses

Question 16

A media company is deploying a generative image model that lets marketing staff create promotional artwork from text prompts. Leadership is concerned that the model could occasionally produce harmful, violent, or otherwise offensive imagery from ambiguous prompts. Which responsible AI principle is most directly addressed by adding content filters that block such outputs?

Accepted Answer

Reliability and safety

Answer

Transparency

Answer

Inclusiveness

Answer

Accountability

Question 17

An e-commerce company wants to automatically detect when two product listings submitted by different sellers describe essentially the same item, even when the wording differs (for example, 'wireless noise-cancelling headphones' vs 'cordless sound-blocking earphones'). Which AI technique is best suited to measure how closely the meaning of two text passages matches?

Accepted Answer

Semantic similarity using text embeddings to compare meaning

Answer

Keyword extraction to pull the most frequent words from each listing

Answer

Sentiment analysis to score the tone of each listing

Answer

Named entity recognition to detect product names in each listing

Question 18

A medical imaging startup is building an AI system that flags suspicious lesions in X-ray images. The team decides that whenever the model's prediction confidence falls below a set threshold, the case is automatically routed to a human radiologist instead of being auto-classified. Which responsible AI principle does this design choice primarily support?

Accepted Answer

Reliability and safety

Answer

Inclusiveness

Answer

Privacy and security

Answer

Transparency

Question 19

You are building a single-agent travel-booking assistant in Microsoft Foundry. The agent must be able to check real-time seat availability by calling your company's existing internal REST API during a conversation. Which capability should you configure on the agent to enable this?

Accepted Answer

Add a custom function (tool) that the agent can invoke to call the internal REST API

Answer

Increase the model's max tokens so it can store the API responses in the prompt

Answer

Lower the temperature parameter so the agent produces deterministic API calls

Answer

Deploy a larger multimodal model to interpret the API documentation automatically

Question 20

A logistics company wants to build a lightweight information-extraction app that pulls the shipment ID, carrier name, and delivery date from thousands of scanned bill-of-lading documents. Before running any documents through Azure Content Understanding in Foundry Tools, what must the team define so the service knows which pieces of information to return?

Accepted Answer

A field schema (analyzer) that specifies the named fields to extract from each document

Answer

A prompt template with a system message that describes the company's tone of voice

Answer

A fine-tuning dataset of thousands of labeled example completions for a chat model

Answer

An embedding index that stores vector representations of every document page

Question 21

A media company records weekly podcast episodes and wants to automatically build a searchable catalog. For each audio file, they need to extract structured data such as the episode title, list of guest names, and the main topics discussed. They want to use a prebuilt Foundry capability designed to pull structured information from unstructured content, including audio, without training a custom machine learning model. Which Foundry Tool should they use?

Accepted Answer

Azure Content Understanding

Answer

Azure Speech (text-to-speech synthesis)

Answer

A fine-tuned generative chat model

Answer

Azure AI Vision image analysis

Question 22

A logistics company receives a mix of shipping documents each day: some are scanned bills of lading, others are packing slips, and some are delivery photos taken by drivers. The team wants to build a lightweight information-extraction app in Azure AI Foundry that pulls structured fields (such as tracking number, weight, and destination) from all of these inputs. Which Foundry Tools capability should they use as the foundation of this app?

Accepted Answer

Azure Content Understanding, defining an analyzer schema with the fields to extract

Answer

Azure Speech, configuring real-time transcription of the documents

Answer

A generative image model, producing new versions of each document

Answer

Text summarization, condensing each document into a short abstract

Question 23

A legal operations team receives hundreds of scanned PDF contracts each week. They need to build a lightweight app that automatically pulls specific fields — such as the counterparty name, effective date, and total contract value — into a structured format for their database. Which Azure AI Foundry capability should they use to define the fields and extract this information from the documents?

Accepted Answer

Azure Content Understanding in Foundry Tools

Answer

Azure Speech in Foundry Tools

Answer

A deployed text-generation model in the Foundry playground

Answer

A multimodal image-generation model

Question 24

A consumer electronics company receives thousands of photographed warranty registration cards from customers. Each card contains printed and handwritten fields such as product serial number, purchase date, and store name. The team wants to build a lightweight app in Microsoft Foundry that automatically pulls these specific field values into structured data for their database. Which Foundry capability should they use?

Accepted Answer

Azure Content Understanding to extract defined fields from the card images

Answer

Azure Speech to transcribe audio descriptions of each card

Answer

A text-generation model with a system prompt that summarizes card contents

Answer

An image-generation model to recreate the cards as clean digital templates

Question 25

A media company records training webinars and needs an automated pipeline that extracts the presenter's spoken talking points, on-screen slide titles, and key visual moments from each recorded video file. Which Azure AI capability in Foundry Tools should the team use to build this single information-extraction solution?

Accepted Answer

Azure Content Understanding, configured with a video analyzer schema

Answer

Azure Speech real-time transcription only

Answer

A text-analysis app using key phrase extraction

Answer

An image-classification model applied to each frame

Question 26

A media company wants to build a lightweight app that automatically extracts spoken dialogue, on-screen text, and key visual moments from uploaded promotional videos so editors can quickly find relevant clips. The team wants to use a single Azure AI service in Foundry Tools that handles all these modalities from the video files. Which service should they use?

Accepted Answer

Azure Content Understanding

Answer

Azure Speech (speech-to-text only)

Answer

Azure AI Vision image classification

Answer

A text-only generative model deployment

Question 27

A logistics company wants to build a lightweight information-extraction app using Azure Content Understanding in Foundry Tools to pull specific fields (shipment ID, destination, weight) from thousands of scanned bill-of-lading documents. Before the app can process any documents, what must the team do first in the Content Understanding workflow?

Accepted Answer

Create an analyzer that defines a schema specifying the named fields to extract

Answer

Train a custom neural voice to read the extracted values aloud

Answer

Deploy a text-to-image generative model to visualize the shipments

Answer

Configure a top-p sampling value to control extraction randomness

Question 28

A public transit authority is building a self-service kiosk. Riders type their destination question into a screen, and the kiosk must read the answer aloud through a speaker so that visually impaired riders can hear it. Which AI workload capability should the developers use to produce the spoken output?

Accepted Answer

Speech synthesis (text-to-speech)

Answer

Speech recognition (speech-to-text)

Answer

Speech translation

Answer

Text summarization

Question 29

A media monitoring company wants to automatically determine whether social media posts about their client express a positive, negative, or neutral opinion, and produce a score reflecting that tone. Which text-analysis technique should they use?

Accepted Answer

Sentiment analysis

Answer

Entity detection

Answer

Key phrase extraction

Answer

Language detection

Question 30

A medical transcription team uses Azure Speech in Foundry Tools to convert dictated notes into text. The default speech-to-text model frequently misrecognizes specialized drug names and clinical terminology, producing inaccurate transcripts. The team wants to improve recognition accuracy for this vocabulary without abandoning Azure Speech. What should they do?

Accepted Answer

Create a Custom Speech model trained with domain-specific vocabulary and sample audio, then deploy it as the recognition endpoint

Answer

Increase the audio sample rate of the recordings so the default model captures more detail

Answer

Switch from speech-to-text to text-to-speech synthesis for the dictated notes

Answer

Enable profanity filtering so unrecognized clinical terms are removed from the transcript

Question 31

A software company receives thousands of free-text customer comments each week. Analysts want an AI solution that, for each comment, produces a single overall rating indicating whether the comment expresses a positive, neutral, or negative opinion so they can track satisfaction trends over time. Which text-analysis technique should they use?

Accepted Answer

Sentiment analysis

Answer

Key phrase extraction

Answer

Entity recognition

Answer

Summarization

Question 32

A retail company collects thousands of written product reviews each week. The analytics team wants to automatically determine whether each review expresses a positive, neutral, or negative opinion so they can track customer satisfaction trends over time. Which text-analysis technique should they use?

Accepted Answer

Sentiment analysis

Answer

Entity detection

Answer

Keyword extraction

Answer

Summarization

Question 33

A bank is deploying a generative AI chatbot that lets customers ask questions about their accounts. During design reviews, the team wants to ensure the system never exposes one customer's personal account details to a different customer and complies with data protection regulations. Which responsible AI principle is the team primarily addressing?

Accepted Answer

Privacy and security

Answer

Transparency

Answer

Inclusiveness

Answer

Reliability and safety

Question 34

A developer at a marketing agency wants to quickly test how a large language model responds to different prompts before writing any application code. They have an Azure AI Foundry project and need the fastest way to deploy a model and experiment with prompts interactively. What should they do first in the Azure AI Foundry portal?

Accepted Answer

Deploy the model from the model catalog, then use the chat playground in the portal to send prompts and view responses

Answer

Install the Foundry SDK locally and write a Python script that authenticates and calls the model endpoint

Answer

Configure an Azure Content Understanding analyzer to extract fields from sample documents

Answer

Create a multi-agent orchestration and connect external tools before sending any prompts

Question 35

A financial services company is deploying a generative AI model in Azure AI Foundry. Regulations require that all customer data processed by the model must remain within the company's own country. Which deployment configuration decision most directly addresses this requirement?

Accepted Answer

Selecting the Azure region where the model is deployed

Answer

Increasing the maximum tokens allowed per response

Answer

Lowering the temperature value for more deterministic output

Answer

Choosing a smaller model to reduce inference latency

Question 36

A field-service company wants a lightweight app where technicians upload a photo of a wiring diagram and ask a typed question such as "Which terminal connects to the ground wire?" The app must reason over the image and the text together and return a written answer. Which model capability should you deploy in Azure AI Foundry to power this app?

Accepted Answer

A multimodal model that accepts both image and text inputs in the same prompt

Answer

A text-only chat completion model with a large context window

Answer

An image-generation model that produces new images from text prompts

Answer

A speech-to-text model that transcribes spoken audio into text

Question 37

A marketing team wants to automatically scan thousands of user-uploaded social media photos to find images that contain their company's logo, so they can measure brand exposure. Which AI workload best fits this requirement?

Accepted Answer

Computer vision (image analysis)

Answer

Sentiment analysis

Answer

Speech recognition

Answer

Text summarization

Question 38

A manufacturing company wants to install cameras on its assembly line that automatically inspect each product and flag units with visible scratches, dents, or misaligned components before they are packaged. Which type of AI workload best fits this requirement?

Accepted Answer

Computer vision

Answer

Natural language processing

Answer

Speech recognition

Answer

Generative AI

Question 39

A bank wants a system that continuously monitors credit card transactions and automatically flags unusual spending patterns that deviate significantly from a customer's normal behavior, so investigators can review potential fraud. Which type of AI workload best fits this requirement?

Accepted Answer

Anomaly detection

Answer

Sentiment analysis

Answer

Image classification

Answer

Speech synthesis

Question 40

A social media startup wants an AI service that automatically scans user-submitted text and images to flag hate speech, violence, and sexually explicit material before it is published on their platform. Which type of AI workload best fits this requirement?

Accepted Answer

Content moderation

Answer

Sentiment analysis

Answer

Optical character recognition (OCR)

Answer

Entity recognition

Question 41

A global e-commerce company receives customer messages through its website contact form. Before routing each message to the correct regional support team, the company needs to automatically determine which language each message is written in. Which text-analysis capability should the company use?

Accepted Answer

Language detection

Answer

Sentiment analysis

Answer

Key phrase extraction

Answer

Named entity recognition

Question 42

A media monitoring company wants to automatically scan thousands of news articles and pull out every company name, person, and geographic location mentioned in each article so analysts can build relationship maps. Which text-analysis technique should they use?

Accepted Answer

Named entity recognition (entity detection)

Answer

Sentiment analysis

Answer

Text summarization

Answer

Language detection

Question 43

A logistics company receives a mix of scanned shipping manifests, PDF bills of lading, and photographed delivery receipts. A developer needs to build a lightweight app that automatically extracts structured fields (such as tracking numbers, dates, and addresses) from all of these document types with minimal custom model training. Which Azure AI Foundry capability should the developer use?

Accepted Answer

Azure Content Understanding in Foundry Tools

Answer

Azure Speech in Foundry Tools

Answer

A generative image model deployed in the Foundry portal

Answer

An embedding model for semantic search

Question 44

A logistics company scans thousands of paper delivery receipts. They need a solution that returns the printed text on each receipt as machine-readable characters, without requiring the system to understand the meaning of specific fields. Which computer vision capability should they use?

Accepted Answer

Optical character recognition (OCR) to read the printed text into digital characters

Answer

Image classification to assign a single category label to each receipt

Answer

Object detection to draw bounding boxes around every physical item shown

Answer

An image-generation model to recreate a cleaner version of each receipt

Question 45

A software company receives thousands of free-text customer support tickets each day. The support manager wants to automatically identify the most important terms and topics in each ticket so agents can quickly understand what a ticket is about without reading the full text. Which text-analysis technique should be applied?

Accepted Answer

Key phrase extraction

Answer

Sentiment analysis

Answer

Language detection

Answer

Speech synthesis

Question 46

A logistics company receives thousands of scanned paper shipping manifests. They need to automatically read the printed text and extract structured fields such as tracking numbers and delivery dates from the images. Which technique should they use to convert the text content of these images into machine-readable data?

Accepted Answer

Optical character recognition (OCR)

Answer

Sentiment analysis

Answer

Image classification

Answer

Speech synthesis

Question 47

A retail company is building a lightweight vision app in Azure AI Foundry. Employees photograph paper receipts with their phones, and the app must read the printed text from these images so it can be stored and searched. Which capability should the app use to convert the printed text in the photos into machine-readable text?

Accepted Answer

Optical character recognition (OCR)

Answer

Image classification

Answer

Text-to-speech synthesis

Answer

Object detection

Question 48

A legal services company uses a large language model to draft contract clauses. The model produces accurate content, but the wording rarely matches the firm's very specific house style and preferred legal phrasing. The team has hundreds of example clauses written in the correct style. Which approach should they use to make the model consistently produce output in the firm's style?

Accepted Answer

Fine-tune the model using the collection of example clauses written in the firm's style

Answer

Increase the temperature parameter to encourage more creative wording

Answer

Reduce the max tokens value so responses are shorter and more focused

Answer

Switch to a smaller model to lower the response latency

Question 49

A developer building a single-agent solution with the Foundry SDK has already created the agent with instructions and connected a Code Interpreter tool. They now want the agent to analyze a specific CSV file the user provides during a conversation and return a chart. Which sequence of SDK steps correctly runs this request?

Accepted Answer

Create a thread, add a user message that references the uploaded file, create a run on the agent, then poll the run and retrieve the resulting messages.

Answer

Redeploy the agent with the file embedded in its instructions, then call the model endpoint directly.

Answer

Delete the existing agent and recreate it with the CSV file attached to the system prompt before sending any message.

Answer

Send the file to the Content Understanding analyzer first, then pass the extracted text to a chat completion call without using the agent.

Question 50

You are building a single-agent solution in Microsoft Foundry for an HR team. The agent must always answer employees in a friendly, formal tone and must never provide legal advice. It also needs to look up answers in the company's uploaded benefits PDF documents. Where should you specify the tone and the restriction against giving legal advice?

Accepted Answer

In the agent's instructions

Answer

By adding a File Search tool to the agent

Answer

By adding a Code Interpreter tool to the agent

Answer

In each individual user message sent to the agent

Question 51

You are building a single-agent solution in Microsoft Foundry that helps analysts explore uploaded sales spreadsheets. Users will ask the agent to calculate monthly revenue trends, compute statistics, and produce charts from the data in the files they upload. Which capability should you add to the agent so it can perform these calculations and generate the charts?

Accepted Answer

The Code Interpreter tool

Answer

The File Search tool

Answer

A detailed system prompt describing statistical formulas

Answer

The Bing web search grounding tool

Question 52

You are building a single-agent solution in Microsoft Foundry for a finance team. Users will upload monthly CSV files and ask the agent to calculate totals, compute month-over-month growth percentages, and produce charts from the numbers in those files. Which tool should you add to the agent to satisfy these requirements?

Accepted Answer

The Code Interpreter tool

Answer

The File Search tool

Answer

The Web Search (Grounding with Bing) tool

Answer

A custom function tool that calls an external calculator API

Question 53

A developer is building a single agent in Microsoft Foundry for an HR team. The agent must answer employee questions by referencing the company's uploaded PDF employee handbook, and it should also calculate remaining vacation days from numeric data the employee types into the chat. Which combination of tools should the developer attach to the agent?

Accepted Answer

The file search tool for the handbook and the code interpreter tool for the vacation calculations

Answer

The web search tool for the handbook and the code interpreter tool for the vacation calculations

Answer

Only the file search tool, which handles both document lookup and numeric calculation

Answer

Only the code interpreter tool, which can read PDFs and perform calculations

Question 54

You are building a single-agent solution in Microsoft Foundry to help HR staff answer employee questions about vacation policies. Before adding any tools, you want the agent to behave as a friendly HR assistant that only answers HR-related questions and refuses off-topic requests. Which element of the agent definition should you configure to establish this behavior?

Accepted Answer

The agent's instructions

Answer

A code interpreter tool

Answer

The deployment region

Answer

A file search tool connection

Question 55

You are building a single-agent solution in the Azure AI Foundry portal. Before other developers on your team can find and reuse the agent, you must configure a property that gives the agent a human-readable identity and a short summary of what it does. Which agent configuration should you set for this purpose?

Accepted Answer

The agent's name and description

Answer

The temperature and top-p sampling parameters

Answer

The system instructions that define behavior

Answer

The connected tools such as file search

Question 56

You are building a lightweight client for an agent using the Foundry SDK. The agent must remember earlier turns of a multi-turn conversation with a single user so it can answer follow-up questions that reference prior messages. Which SDK concept should you use to maintain the ongoing conversation state across requests?

Accepted Answer

Create a thread and add each user message to it, then run the agent against that same thread for every turn

Answer

Increase the max_tokens value so the model can hold more of the conversation in a single response

Answer

Send the entire conversation as one large system prompt on every request and discard the response history

Answer

Raise the temperature setting so the agent produces more contextually varied answers

Question 57

Your team has just created a single-agent solution in the Microsoft Foundry portal that answers HR benefits questions using a file search tool over policy PDFs. Before your developers write any client code with the Foundry SDK, you want to quickly confirm the agent responds correctly and grounds its answers in the uploaded documents. What is the most efficient way to do this?

Accepted Answer

Interact with the agent directly in the Foundry portal's agent playground and review its responses

Answer

Deploy a lightweight web client with the Foundry SDK and have testers submit questions through it

Answer

Publish the agent to production and monitor real user conversations for accuracy

Answer

Export the agent configuration to a JSON file and manually inspect the file search settings

Question 58

You are building a single-agent solution in Microsoft Foundry that answers customer questions and must return each response as a structured JSON object so a downstream web application can parse the fields reliably. What is the most appropriate way to ensure the agent consistently produces this format?

Accepted Answer

Include instructions in the agent's system prompt that specify the exact JSON schema and provide an example of the expected output

Answer

Increase the temperature parameter so the agent produces more varied responses

Answer

Add a file search tool so the agent can look up the JSON schema in an uploaded document

Answer

Reduce the max tokens setting so the response is short enough to be valid JSON

Question 59

A developer is building a single-agent expense assistant in Microsoft Foundry. Employees will paste a photograph of a paper receipt into the chat and ask the agent to summarize the total and merchant name. The agent must be able to read the printed text visible in the uploaded photo before answering. Which capability should the developer rely on to enable this?

Accepted Answer

Deploy a multimodal model that can accept image input alongside the text prompt

Answer

Add a code interpreter tool so the agent can run Python to parse the receipt

Answer

Enable a web search tool so the agent can look up merchant details online

Answer

Increase the model's max tokens so it can process the larger image payload

Question 60

You are building a single-agent solution in Microsoft Foundry for an internal help desk. The agent must answer employee questions strictly from a set of uploaded company HR policy PDFs and must not pull information from the public internet. Which tool should you attach to the agent?

Accepted Answer

The File Search tool connected to the uploaded HR policy PDFs

Answer

The Grounding with Bing Search (web search) tool

Answer

The Code Interpreter tool

Answer

The Function Calling tool pointing to a public weather API

Question 61

You are building a generative AI app in Microsoft Foundry that classifies incoming customer messages into predefined categories such as 'Billing', 'Technical', and 'General'. During testing you notice the model sometimes invents category names that are not on your approved list. Without changing the deployed model, what is the most effective way to improve the consistency of the output?

Accepted Answer

Include a few example messages with their correct category labels directly in the prompt (few-shot prompting).

Answer

Increase the temperature parameter so the model explores more possible categories.

Answer

Deploy a larger model in a different region to gain more processing capacity.

Answer

Raise the max tokens limit so the model has room to describe each category in detail.

Question 62

A content team uses a deployed multimodal model in the Foundry portal to describe product photos. They now want to test the model's behavior on a single new image quickly, adjusting the prompt wording several times before writing any code. Which approach in Microsoft Foundry lets them do this fastest?

Accepted Answer

Use the model playground in the Foundry portal to upload the image and iterate on the prompt interactively

Answer

Write a lightweight vision client with the Foundry SDK and rerun the script after each prompt edit

Answer

Create a single-agent solution with a file search tool that indexes the image

Answer

Configure an Azure Content Understanding analyzer with an image field schema

Question 63

A retail team wants a lightweight vision app built in Foundry that accepts two product shelf photos in the same prompt and returns which shelf appears better stocked. They ask the developer how to pass both images to the deployed multimodal model. What is the correct approach?

Accepted Answer

Include both images as separate image content items within a single user message alongside the text instruction.

Answer

Send two separate chat completion requests, one per image, and merge the responses in the app code.

Answer

Concatenate the two images into one file first, because a multimodal model can only accept a single image per request.

Answer

Place the images in the system message so the model treats them as fixed instructions.

Question 64

A logistics company wants a lightweight vision app built in Azure AI Foundry. Warehouse workers will take photos of arriving packages on their phones, and the app must automatically read the shipping label text visible in each photo and return it as plain text that can be pasted into a tracking system. The team wants to send the image directly to a deployed model in a single prompt without training a custom model first. Which approach should they use?

Accepted Answer

Deploy a multimodal model and include the package photo as image input in the prompt, asking the model to read and return the label text

Answer

Deploy a text-only generative model and paste a written description of the label into the prompt

Answer

Use Azure Speech in Foundry Tools to transcribe the label from the photo

Answer

Train a custom image classification model to sort packages into categories before extracting text

Exam domain	Exam weight	Practice
Identify AI concepts and capabilities	43%	Practice this topic
Implement AI solutions by using Microsoft Foundry	57%	Practice this topic

Microsoft Azure AI Fundamentals (AI-901) practice exam & study guide

Start studying AI-901

All AI-901 study resources

AI-901 exam domains

Sample AI-901 questions

Key AI-901 terms

AI-901 frequently asked questions

Identify AI concepts and capabilities (43%)

Implement AI solutions by using Microsoft Foundry (57%)