Free AI-900 Practice Questions – 2026 Updated

The chit-chat feature in Azure AI Language's Question Answering (the successor to QnA Maker) is specifically designed to add a conversational layer to a bot. It provides pre-built datasets for common conversational queries, including greetings, small talk, and other user-friendly interactions. By selecting a personality, such as "Professional," you can easily incorporate these responses without having to create them manually. This directly addresses the requirement to make the bot more user-friendly with greetings and other common conversational replies, separate from the core FAQ knowledge base.

A. Increasing the confidence threshold makes the bot stricter about when it provides an answer from the knowledge base; it does not add new conversational content.

B. Active learning is a feature that suggests new question variations to improve the accuracy of the existing knowledge base, not add general conversational greetings.

C. Multi-turn questions are used to create follow-up prompts that guide a user to a specific, refined answer, which is different from adding general chit-chat.

1. Microsoft Learn

Azure AI Language Documentation. "Add chit-chat to a knowledge base." This document states

"Adding chit-chat to your bot makes it more conversational and engaging... When you build your knowledge base

you can add chit-chat from a pre-defined data set." It also details the different personalities available

including "Professional."

2. Microsoft Learn

Azure AI Language Documentation. "Use follow-up prompts to create multiple turns of a conversation." This document explains that multi-turn conversations are used to "refine a question to get to a final answer

" which is distinct from the goal of adding general greetings.

3. Microsoft Learn

Azure AI Language Documentation. "Active learning suggestions." This source describes active learning as a mechanism to "improve the quality of your knowledge base by suggesting alternative questions... based on user submissions." This confirms it is for improving the core Q&A pairs

not for adding conversational filler.

4. Microsoft Learn

Azure AI Language Documentation. "Confidence score." This page explains that the confidence score indicates the confidence in the answer's accuracy. Adjusting the threshold filters results

it does not add new content like greetings.

Azure AI Document Intelligence (formerly Form Recognizer) is a component of the broader Azure AI Services suite. Azure AI Services are a collection of prebuilt, customizable AI models and services available through APIs. The prebuilt receipt model is a specific feature within Azure AI Document Intelligence that allows developers to automatically extract key information from receipts, such as merchant name, transaction date, and totals, without needing to train a custom model. Therefore, Azure AI Services is the correct umbrella service that provides this capability.

A. Azure Al Computer Vision: This is a distinct service within Azure AI Services focused on general image analysis, such as object detection and optical character recognition (OCR), not specialized document extraction.

B. Azure Machine Learning: This is a platform for building, training, and deploying custom machine learning models, not for consuming prebuilt AI models like the receipt model.

D. Azure Al Custom Vision: This service is used to build and train custom models for image classification and object detection, not for analyzing text and structure in documents like receipts.

1. Microsoft Learn. "What is Azure AI Document Intelligence?" Azure AI services documentation. "Azure AI Document Intelligence is an Azure AI service that lets you build automated data processing software using machine learning technology." This establishes Document Intelligence as part of Azure AI Services.

2. Microsoft Learn. "Prebuilt receipt model - Document Intelligence." Azure AI services documentation. This document details the prebuilt receipt model as a feature of the Document Intelligence service.

3. Microsoft Learn. "What are Azure AI services?" Azure AI services documentation. This page lists the portfolio of services

including "Azure AI Document Intelligence (formerly Form Recognizer)" under the main Azure AI Services category.

4. Microsoft Learn. "AI-900: Describe the features of Azure AI Document Intelligence." Study guide for Exam AI-900: Microsoft Azure AI Fundamentals. This official study guide categorizes Document Intelligence as one of the Azure AI Services that provides prebuilt models for document analysis.

Grounding is a prompt engineering technique where a generative AI model is provided with a specific, relevant set of information (the "grounding data") to use as a source of truth for its response. This technique constrains the model to the provided context, improving the factual accuracy and relevance of its output. This is a core principle behind patterns like Retrieval-Augmented Generation (RAG), where the model retrieves information from a knowledge base to "ground" its answer, rather than relying solely on its general training data.

A. providing examples: This technique, known as few-shot learning, shows the model the desired format or style of response through examples, rather than providing a corpus of factual data for a specific query.

B. fine-tuning: This is a process of retraining a model on a specific dataset to adapt its internal parameters (weights). It is not a prompt engineering technique, which works with the model as-is.

D. system messages: A system message sets the overall persona, tone, and behavioral rules for the AI model for an entire conversation, not the specific contextual data for a single response.

1. Microsoft Learn

"Introduction to prompt engineering": Under the "Common prompt engineering techniques" section

it states

"Grounding is the technique of providing the model with information to use in its response. For example

you might provide a text document and prompt the model to use it to answer questions."

2. Microsoft Learn

"Retrieval Augmented Generation (RAG) in Azure AI Studio": In the "What is Retrieval Augmented Generation (RAG)?" section

the documentation explains that RAG improves model performance by "providing it with a source of information as a foundation

" which is the definition of grounding the model with your own data.

3. Microsoft Learn

"AI-900: Describe considerations for generative AI": In the "Describe prompt engineering" unit

the concept of providing context is central. It explains

"The prompt can include contextual information

such as articles or documents that the model can use to respond." This describes the grounding technique.

The process of generating an audio file from a text script is known as speech synthesis, commonly referred to as text-to-speech (TTS). The scenario requires converting written scripts into spoken narration for videos. The Azure AI Speech service provides this exact capability, enabling the creation of natural-sounding synthesized speech from text input. This directly addresses the need to generate an audio file for each video's script.

A. speech recognition: This converts spoken audio into text (speech-to-text), which is the opposite of the required task.

B. language modeling: This is primarily used for understanding and generating human-like text, not for converting text into audio.

D. translation: This converts text from a source language to a target language, not from text to an audio format.

1. Microsoft Learn

"What is text-to-speech?": "Text-to-speech (TTS)

also known as speech synthesis

is the technology that converts text into audible speech. You can use text-to-speech to create synthesized voices for your applications and services." This source directly defines speech synthesis as the correct workload.

2. Microsoft Learn

"AI-900: Describe features of speech workloads on Azure": In the "Speech synthesis (text-to-speech)" section

it states

"Speech synthesis is the capability of software to generate audible speech from text." This aligns perfectly with the question's requirement to generate narration from a script.

3. Microsoft Learn

"What is speech recognition?": "Speech recognition is the process of transcribing spoken audio to text." This reference confirms that speech recognition is the inverse of the required function.

Fine-tuning is the process of customizing a pre-trained foundation model by continuing the training process using your own specific dataset. This involves uploading a set of training examples (often prompt-completion pairs) to Azure OpenAI. The service then uses this data to update the model's weights, adapting its behavior and knowledge to better suit a particular task or domain. This results in a new, custom model that is more specialized than the original base model.

A. completion: A completion is the text generated by the model in response to a prompt; it is the output, not the customization process.

B. grounding: Grounding involves providing the model with specific, timely data at the time of inference (e.g., via RAG), but it does not permanently alter the model's underlying parameters.

D. prompt engineering: This is the technique of carefully crafting input prompts to guide the model's output, which does not involve uploading data to retrain the model.

1. Microsoft Learn

Azure OpenAI Service Documentation. "Customize a model with fine-tuning". This document states

"Fine-tuning improves on few-shot learning by training the model on many more examples than can fit in the prompt. To fine-tune a model

you prepare a dataset of training examples...". This confirms that fine-tuning is the process of training on your own data.

2. Microsoft Learn

AI-900 Learning Path. "Use your own data with Azure OpenAI Service". This module distinguishes between different methods. In the "Understand fine-tuning" section

it clarifies

"Fine-tuning is a technique you can use to customize a model by training it on your own data."

3. Microsoft Learn

Azure OpenAI Service Documentation. "What is fine-tuning for Azure OpenAI Service?". The overview section states

"Fine-tuning is a feature of Azure OpenAI Service that allows you to customize a base model for your specific application... To fine-tune a model

you need to provide a set of training examples...".

Speech recognition, also known as speech-to-text, is the process of transcribing spoken audio into written text. Providing closed captions for videos (B) requires converting the spoken dialogue into text to be displayed on screen. Similarly, creating a transcript of a call or meeting (D) involves converting the spoken words of participants into a written document. Both are core applications of speech recognition technology.

A. An in-car system that reads text messages aloud is an example of text-to-speech (speech synthesis), which converts written text into spoken audio, the opposite of speech recognition.

C. An automated public address system converts pre-written or dynamically generated text into spoken announcements. This is a text-to-speech application, not speech recognition.

1. Microsoft Learn

AI-900: Explore speech capabilities. In the "Transcribe speech input to text" unit

it states

"Speech-to-text is used to transcribe spoken audio into text. Common uses for speech-to-text include creating transcripts of meetings..." This directly supports option D.

2. Microsoft Azure AI Speech service documentation

"What is speech-to-text?". Under the "Key features" section

"Real-time speech-to-text" is described for use cases such as "live captioning

" which directly supports option B.

3. Microsoft Azure AI Speech service documentation

"What is text-to-speech?". This document explains that text-to-speech "enables your applications

tools

or devices to convert text into humanlike synthesized speech." This confirms that scenarios A and C

which involve reading text aloud

are text-to-speech functionalities.

Anomaly detection is a machine learning technique focused on identifying data points, events, or observations that deviate from a dataset's normal behavior. These non-conforming instances are often referred to as anomalies, outliers, or exceptions. A primary application of this technique is to analyze time-series data to detect unexpected spikes, dips, or changes in behavioral patterns. This is crucial for scenarios like fraud detection in financial transactions, monitoring equipment for predictive maintenance, or identifying network security intrusions. The core function is to flag unusual events that warrant further investigation.

A. This describes Natural Language Processing (NLP), which deals with the interaction between computers and human language, not identifying general data anomalies.

B. This describes object detection, a computer vision task that identifies and locates objects within an image, not a general technique for finding unusual data patterns.

D. This describes image classification, a computer vision task that assigns a label to an entire image based on its content, not anomaly detection.

1. Microsoft Learn. (2024). Introduction to anomaly detection. AI-900: Microsoft Azure AI Fundamentals. "Anomaly detection is a machine learning technique that is used to identify unusual patterns or outliers in data." Retrieved from https://learn.microsoft.com/en-us/training/modules/explore-anomaly-detection/1-introduction

2. Microsoft Learn. (2024). What is the Anomaly Detector service?. Azure AI services documentation. "The Anomaly Detector service enables you to monitor and detect abnormalities in your time series data... Anomaly Detector ingests time-series data of any kind and selects the best anomaly detection algorithm for your data to ensure high accuracy." Retrieved from https://learn.microsoft.com/en-us/azure/ai-services/anomaly-detector/overview

3. Microsoft Learn. (2024). Describe machine learning. AI-900: Microsoft Azure AI Fundamentals. This module distinguishes between different machine learning workloads

including anomaly detection

regression

classification

and clustering

clarifying their distinct purposes. Retrieved from https://learn.microsoft.com/en-us/training/modules/describe-machine-learning/

The requirement is to create an original image using Azure OpenAI. The DALL-E model family is specifically designed for this purpose, generating novel images from natural language text descriptions (prompts). Within the Azure AI ecosystem, you would select and deploy a DALL-E model from the model catalog (found in Azure AI Studio) and then provide it with instructions, such as "a peaceful rural scene with a red barn and rolling green hills at sunset," to generate the required image.

A. GPT-3.5 Turbo is a large language model designed for generating and understanding text. It cannot create images.

B. Searching on Microsoft Bing retrieves existing images from the internet. This does not create an original image and may involve copyright issues.

C. GitHub Copilot is an AI assistant for software developers that suggests code and entire functions; it is not a tool for image generation.

1. Microsoft Learn

Azure AI Services Documentation. "Azure OpenAI Service models." This document explicitly categorizes DALL-E models for "Image generation" and GPT-3.5 Turbo models for "Text generation

" clearly distinguishing their functions.

2. Microsoft Learn

AI-900 Courseware. "Module: Explore generative AI on Microsoft Azure

" Unit 4: "Explore image generation." This unit details how generative AI can be used to create images and specifically mentions the DALL-E model as the tool for this task within Azure.

3. Microsoft Learn

Azure AI Services Documentation. "Quickstart: Get started generating images with Azure OpenAI Service." This guide provides a step-by-step tutorial

stating in the introduction

"This quickstart shows you how to use the DALL-E... models to generate images with the Azure OpenAI Service."

The principle of inclusiveness in Microsoft's responsible AI framework is centered on designing AI systems that empower everyone and engage all people. This explicitly includes ensuring that systems are accessible and usable by individuals with diverse abilities, including those with hearing, visual, or other impairments. The scenario's goal of creating a system that serves this wide range of users directly aligns with the core tenet of inclusiveness.

A. fairness: This principle focuses on ensuring AI systems treat all people equitably and mitigate biases related to factors like gender, ethnicity, or other demographics.

C. reliability and safety: This principle ensures that AI systems operate dependably, securely, and consistently as intended, without causing unintended harm.

D. accountability: This principle establishes that the people who design and deploy AI systems are answerable for their operation and impact, emphasizing human oversight.

---

1. Microsoft Learn

"Describe principles of responsible AI" (AI-900 Learning Path).

Section: Inclusiveness. The documentation states

"AI systems should empower everyone and engage people. For people with a hearing

visual

or other impairment

Microsoft AI products and services provide features like real-time captioning

text analytics

and computer vision technology..." This directly supports the correct answer by linking inclusiveness to designing for impairments.

2. Microsoft

"Our approach to responsible AI at Microsoft."

Section: Our principles > Inclusiveness. This official page defines the principle: "Microsoft is committed to creating responsible AI

with inclusiveness as a key principle. This means that AI systems must be designed to be inclusive of all people

regardless of their background

abilities

or identity."

3. Microsoft

"The Responsible AI Standard

v2."

Page 10

Section: 3. Microsoft’s Responsible AI Principles > 3.4 Inclusiveness. The standard details that inclusiveness involves "ensuring that AI systems are designed to be accessible to people with disabilities" and "avoiding the creation or reinforcement of barriers that exclude people from participating in society." This provides a formal definition that matches the question's scenario.

Content filtering is the specific safety system implemented in generative AI services, such as Azure OpenAI, to identify and manage harmful content. This system analyzes both user inputs (prompts) and the AI's generated outputs (responses) in near-real time. It uses a set of classification models to detect and categorize potentially harmful content, including hate speech, violence, and other inappropriate material. When a hateful response is identified, the system can block it from being returned to the user, thereby mitigating the risk of exposing users to harmful content.

A. prompt engineering: This is the practice of designing inputs to guide the AI's output. While it can reduce the likelihood of harmful responses, it is not the mechanism that identifies them.

B. abuse monitoring: This generally refers to tracking service usage patterns (e.g., API call rates) to prevent system misuse, rather than analyzing the specific content of AI responses.

D. fine-tuning: This is a process of further training a model on a custom dataset to specialize its behavior. It is a model development technique, not a real-time safety system for identifying harmful output.

1. Microsoft Learn

"Azure OpenAI Service content filtering." This document states

"Azure OpenAI Service includes a content filtering system that works alongside core models. The system works by running both the prompt and completion through an ensemble of classification models aimed at detecting and preventing the output of harmful content." It explicitly lists Hate as a primary content category that is filtered.

2. Microsoft Learn

"Responsible AI practices for Azure OpenAI models." Under the section "Mitigating harms

" it details that "Content filters are a key component of our responsible AI strategy... to detect harmful content and prevent it from being generated by the service."

3. Microsoft Learn

"Introduction to responsible generative AI." In the "Safety" section

it explains

"Safety systems for generative AI models often include... Content filters to detect and prevent the generation of harmful or inappropriate content." This confirms that content filtering is the designated tool for this purpose.

The scenario requires converting text from a source language (English) to a target language based on the user's location. This task is the definition of machine translation. The Azure AI Language service offers text translation capabilities that take source text and return the translated text in one or more target languages, directly addressing the requirement to present reviews to users in their respective languages.

B. language modeling: This is a foundational NLP task that predicts the probability of a sequence of words; it does not translate text between languages.

C. key phrase extraction: This feature identifies the main points or topics in a text. It analyzes content but does not convert it to another language.

D. speech recognition: This workload converts spoken audio into written text. The input in this scenario is customer reviews, which are already in text format.

1. Microsoft Learn

AI-900: Microsoft Azure AI Fundamentals

"Explore natural language processing"

Unit: "Understand text and speech translation": This unit explicitly defines the workload. It states

"Text translation is the task of translating text from one language to another... You can build solutions that use machine translation to overcome language barriers." This directly supports the choice of translation for the described scenario.

2. Microsoft Learn

AI-900: Microsoft Azure AI Fundamentals

"Explore natural language processing"

Unit: "Understand language": This section describes key phrase extraction as a feature to "identify important words and phrases in the text that suggest its main points

" confirming it is for analysis

not translation.

3. Microsoft Learn

AI-900: Microsoft Azure AI Fundamentals

"Explore natural language processing"

Unit: "Understand speech": This unit defines speech recognition (also known as speech-to-text) as "the ability to recognize and transcribe spoken language

" which is not applicable to the text-based reviews in the question.

The core requirement is to determine the location of cars within an image. Object detection is the specific computer vision technique designed to identify the presence, class, and location of multiple objects in an image. It returns bounding box coordinates for each detected object (in this case, each car), which is the necessary information to then estimate the distance between them.

A. optical character recognition (OCR) is used to detect and extract printed or handwritten text from images, which is not relevant to locating cars.

C. image classification assigns a single label to an entire image (e.g., "an image of a road with cars") but does not identify the location of individual objects.

D. face detection is a specialized form of object detection that is trained specifically to locate human faces, not vehicles.

1. Microsoft Learn. (2024). Explore computer vision in Microsoft Azure. AI-900: Microsoft Azure AI Fundamentals course. In the "Object detection" section

it states

"Object detection is a form of computer vision in which a model is trained to classify individual objects within an image

and indicate their location with a bounding box." This directly supports the correct answer.

2. Microsoft Learn. (2024). What is Computer Vision?. Azure AI Vision documentation. Under the "Object detection" section

it clarifies

"Object detection is similar to tagging

but the API returns the bounding box coordinates (in pixels) for each object found." This confirms that location data is a key output.

3. Microsoft Learn. (2024). Explore computer vision in Microsoft Azure. AI-900: Microsoft Azure AI Fundamentals course. The "Image classification" section contrasts this by explaining it answers the question

"'what is this image?'"

without mentioning object location

confirming why option C is incorrect.

The question asks to predict a continuous numerical value ("how many hours"). This is the definition of a regression task. Regression models are used in supervised learning to predict quantitative outcomes. In contrast, classification is used to predict a discrete category or class (e.g., "will work overtime" vs. "will not work overtime"). Clustering is an unsupervised learning method used to group unlabeled data based on similarities, which is not applicable to this predictive scenario.

Coursera. (n.d.). Microsoft Azure Machine Learning for Data Scientists. "Create a Regression Model with Azure Machine Learning Designer" module. States: "Regression is a supervised machine learning technique used to predict numeric values." (Source: https"//www.coursera.org/learn/microsoft-azure-machine-learning-for-data-scientist)

Microsoft. (n.d.). How to select algorithms for Azure Machine Learning. Microsoft Learn. Identifies "Regression" as a family of classic, traditional machine learning tasks distinct from classification. (Source: https"//learn.microsoft.com/en-us/azure/machine-learning/how-to-select-algorithms?view=azureml-api-1)

Abe, N., et al. (2023). Monetary Flood Damage Prediction Based On Machine Learning Models. Journal of Student Research. "Classification models differ from regression models because rather than predicting a single numerical value, they predict whether a sample falls within a category." (Source: https"//www.jsr.org/hs/index.php/path/article/download/6588/3004/43383, Page 14)

MIT OpenCourseWare. (n.d.). 6.390: Introduction to Machine Learning, Regression Lab. Provides an example of regression: "In this problem we will use linear regression to study the relationship between variables... we will try to predict the attribute Percent_Person_Obesity..." (Source: https"//ocw.mit.edu/courses/res-tll-008-social-and-ethical-responsibilities-of-computing-serc/mitres-tll008s23_6039_regression.pdf, Page 3)

Microsoft. (n.d.). K-Means Clustering: Component Reference. Microsoft Learn. Defines clustering, the other incorrect option: "K-means is one of the simplest and the best known unsupervised learning algorithms... clustering uses iterative techniques to group cases in a dataset..." (Source: https"//learn.microsoft.com/en-us/azure/machine-learning/component-reference/k-means-clustering?view=azureml-api-2)

The Microsoft principle of Reliability and Safety dictates that AI systems must behave dependably and as intended, especially when facing unexpected conditions. Unusual or missing values are classic examples of such conditions. An AI system that fails, produces errors, or behaves unpredictably when given this typeof input is inherently unreliable and potentially unsafe. This principle ensures systems are robust enough to handle data imperfections gracefully. While other principles are important, they are less direct: inclusiveness relates to fairness and bias, privacy and security to data protection, and transparency to explainability.

Microsoft. (2022). Microsoft Responsible AI Standard, v2. "Reliability and Safety" Principle, pp. 11-12.

Quote/Paraphrase: The standard defines this principle as ensuring "AI systems behave safely, reliably, and as intended." A key goal is "robustness," which includes the system's ability to "handle unexpected conditions and failures." Unusual or missing values are a primary form of unexpected conditions that test a system's robustness.

Microsoft. (2024). "Reliability for AI systems." Microsoft Learn documentation, Azure Well-Architected Framework.

Quote/Paraphrase: This documentation outlines that a reliable system must "handle failures gracefully." It emphasizes data validation and managing data quality, stating that "data integrity is crucial" and the system must be resilient to "data corruption" or "missing data," which directly supports this answer.

Hauer, T., et al. (2023). "A Survey on Robustness in Deep Learning: Attacks, Defenses, and Future Directions." ACM Computing Surveys, 55(11), Article 233.

DOI: https://doi.org/10.1145/3564906

Quote/Paraphrase: This survey (Section 3.1) discusses "Data Quality and Preprocessing" as a key challenge for robustness, which is a core component of reliability. It explicitly links "missing values" and "out-of-distribution (OOD)" samples (i.e., unusual values) to a model's robustness, reinforcing that handling these issues is a fundamental aspect of creating reliable AI.

The Azure Machine Learning designer is a graphical, low-code/no-code interface. Its primary paradigm involves users dragging pre-built components, known as modules, onto a visual workspace (the "canvas") and then connecting them to form a pipeline. This visual workflow represents the flow of data and the sequence of operations (like data preparation, training, and scoring) used to build and test a machine learning model. The other options describe different services: "automatically selecting an algorithm" defines Automated ML, and "using a code-first notebook experience" describes using the Azure ML SDK in notebooks.

Microsoft. (n.d.). What is Azure Machine Learning designer? Microsoft Learn. Retrieved October 24, 2025, from https://learn.microsoft.com/en-us/azure/machine-learning/concept-designer

Section: "What is the designer?"

Content: "Azure Machine Learning designer is a drag-and-drop interface... It lets you create models using prebuilt modules on a visual canvas. You can connect the modules to create a pipeline draft...". This directly supports the correct answer.

Microsoft. (n.t.). Tutorial: Predict auto price with the designer. Microsoft Learn. Retrieved October 24, 2025, from https://learn.microsoft.com/en-us/azure/machine-learning/tutorial-designer-automobile-price-train-score

Section: "Introduction"

Content: "The designer provides a visual canvas where you can drag and drop modules to build a pipeline. This low-code environment...". This reinforces the "visual canvas" and "connecting modules" (implied by building a pipeline) concepts.

Microsoft. (n.d.). What is Azure Machine Learning? Microsoft Learn. Retrieved October 24, 2025, from https://learn.microsoft.com/en-us/azure/machine-learning/overview-what-is-azure-machine-learning

Section: "Authoring tools"

Content: This document contrasts the different authoring tools. It describes Automated ML as letting you "train, and test models automatically" (refuting option 3) and Notebooks as providing "a full code-first notebook experience" (refuting option 4). This distinction confirms the designer's unique role as the visual canvas-based tool.

Power Virtual Agents is the correct choice because it is a no-code/low-code platform within the Microsoft Power Platform designed specifically for creating chatbots (virtual agents). It enables non-developers and subject matter experts to build powerful bots using a graphical interface without writing code. Critically, it includes built-in natural language processing (NLP) capabilities to understand user intent, extract entities, and manage the conversation flow, directly matching the description in the question.

• Microsoft Bot Framework is a comprehensive, open-source SDK for developers, requiring coding to build bots, which contradicts the "no-code" requirement.
• Azure Health Bot is a specialized, code-based service built on the Bot Framework, focused on healthcare scenarios, and is not a general-purpose no-code platform.

Official Vendor Documentation (Microsoft Power Platform):

Microsoft Docs: What is Power Virtual Agents? Section: "Key capabilities," defining it as "a low-code tool that you can use to create powerful chatbots." It explicitly mentions that it "uses built-in natural language processing (NLP) models and rich, graphical tooling" for conversational AI. (Reference: Microsoft Learn/Docs, Power Virtual Agents overview section).

Official Vendor Documentation (Microsoft Bot Framework):

Microsoft Docs: Bot Framework Overview: Describes the framework as providing "an SDK, tools, and services" for developers, contrasting it with Power Virtual Agents' no-code approach. (Reference: Microsoft Learn/Docs, Bot Framework SDK documentation).

Academic Publication/Courseware (Contextual Support on No-Code/Low-Code and NLP):

IEEE Software, Vol. 38, Issue 5, 2021: Discussion on the rise of low-code platforms and their integration of complex technologies like AI/NLP to democratize application development for citizen developers. (General Principle: Low-code platforms like Power Virtual Agents abstract the complexity of technologies like NLP).

Official Vendor Documentation (Azure Health Bot):

Microsoft Docs: Azure Health Bot Service: Defines it as a cloud service enabling developers to "build and deploy compliant, AI-powered virtual assistants and chat bots in the healthcare domain." This confirms it is a specialized service for developers, not a general no-code app builder. (Reference: Microsoft Learn/Docs, Azure Health Bot product page).

Anomaly detection (or outlier detection) is a machine learning technique used to identify rare items, events, or observations that deviate significantly from an established "normal" pattern.

1. Forecasting housing prices is a regression problem, which involves predicting a continuous numerical value (price), not identifying an anomaly.
2. Identifying suspicious sign-ins by finding deviations from usual patterns (e.g., new location, unusual time) is a classic application of anomaly detection.
3. Predicting diabetes based on medical history is a classification problem, which involves assigning a data point to a predefined category (e.g., "will develop diabetes" or "will not develop diabetes").

Han, J., Kamber, M., & Pei, J. (2012). Data Mining: Concepts and Techniques (3rd ed.). Morgan Kaufmann Publishers (Elsevier).

Chapter 8, Section 8.1 (Basic Concepts): Defines classification as the task of predicting a categorical class label (e.g., medical diagnosis).

Chapter 8, Section 8.4 (Regression Analysis): Defines regression as the task of modeling and predicting continuous-valued functions (e.g., predicting price).

Chapter 12, Section 12.1 (Outlier Detection): Defines outliers (anomalies) as data objects that deviate significantly from the rest of the data objects. It explicitly lists "intrusion detection" (like suspicious sign-ins) as a key application area.

Chandola, V., Banerjee, A., & Kumar, V. (2009). Anomaly detection: A survey. ACM Computing Surveys (CSUR), 41(3), Article 15.

Section 1 (Introduction): "Anomalies are patterns in data that do not conform to a well defined notion of normal behavior." The survey highlights applications in "network intrusion detection" and "fraud detection," which are analogous to identifying suspicious sign-ins.

DOI: https://doi.org/10.1145/1541880.1541882

Aggarwal, C. C. (2017). Data Mining: The Textbook. Springer.

Chapter 11 (Outlier Analysis), Section 11.1 (Introduction): Defines an outlier as a data point "that is significantly different from the remaining data." It contrasts this with classification, stating that while classification requires predefined labels for all classes (including rare ones), outlier detection traditionally deals with unlabeled data where the "normal" class is implicitly defined.

Chapter 6 (Regression Analysis): Defines regression as the process of learning a functional relationship between variables to predict a numerical target (e.g., stock prices, housing prices).

The Azure AI Translator service is a text-based service. Its primary functions are:

1. Translate: To translate text from a source language into one or more target languages.
2. Detect: To identify the language of a given piece of text.
3. Transliterate: To convert text from one script to another (e.g., Cyrillic to Latin).

The third statement, "transcribe audible speech into text," is the core function of a different service: the Azure AI Speech service. While the Speech service can be used for speech translation, the Translator service itself operates on text input, not audio streams.

Microsoft Corporation. (2024). What is Azure AI Translator? Microsoft Learn.

Reference: This document outlines the service's key capabilities. Under the "What it does" section, it explicitly lists Text translation and Language detection as features. Speech transcription is not listed as a capability of this specific service.

Microsoft Corporation. (2024). Language detection with the Translator service. Microsoft Learn.

Reference: This documentation details the "Detect" operation, confirming that the Translator service is used to "return a language code indicating the language of the supplied text."

Microsoft Corporation. (2024). What is the Speech service? Microsoft Learn.

Reference: This document defines the separate service responsible for transcription. The first capability listed is Speech-to-text, which it defines as "transcribe continuous, real-time, or batch audio into text." This confirms transcription is a function of the Speech service, not the Translator service.

Conversational AI is the specific field of artificial intelligence that enables applications to understand human language and engage in interactive, human-like dialogue. The scenario described—an application interactively answering user-entered questions—is the primary use case for conversational AI technologies, such as chatbots and virtual assistants.

The other options are incorrect:

• Anomaly detection is used to identify unusual patterns or outliers in data, not to answer questions.
• Computer vision is used to process and interpret visual information from images and videos.
• Forecasting is a statistical method used to predict future trends based on historical data.

Microsoft Learn. (n.d.). Conversational language understanding overview. Azure AI services. Retrieved October 24, 2025.

Reference (Correct Answer): In the main overview, the service is described as one that "applies machine-learning intelligence to enable you to build natural language understanding component to be used in an end-to-end conversational application." It lists an "End-to-end conversational bot" as an example that can "process and analyze incoming text in real time to identify the intention of the text" and "perform the desired action," which includes answering questions.

Microsoft Learn. (n.d.). What is Anomaly Detector? Azure AI services. Retrieved October 24, 2025.

Reference (Incorrect Option): The service is defined as enabling "you to monitor and detect anomalies in your time series data," which is unrelated to answering user questions.

Microsoft Learn. (n.d.). What is Azure AI Vision? Azure AI services. Retrieved October 24, 2025.

Reference (Incorrect Option): This service is defined as giving "access to advanced algorithms that process images and return information based on the visual features you're interested in." This relates to visual data, not text-based questions.

Microsoft Learn. (n.d.). Overview of forecasting methods in AutoML. Azure Machine Learning. Retrieved October 24, 2025.

Reference (Incorrect Option): This document describes time series models that "use historical values of the target quantity to make predictions into the future." This is for prediction, not interactive answering.

The Azure Kubernetes Service (AKS) is the most precise and appropriate service for hosting automated ML models in production for high-scale, low-latency inferencing (real-time predictions). AKS is a fully managed container orchestration service that allows for the deployment, scaling, and management of Docker containers. Machine learning models, especially those from Azure Machine Learning, are frequently deployed as containerized web services on AKS because it offers features essential for production ML deployments, such as auto-scaling, rolling updates, and load balancing [Microsoft Official Documentation, "Deploy a model to Azure Kubernetes Service"].

• Azure Data Factory is an ETL (Extract, Transform, Load) service primarily used for data movement and transformation pipelines, not for hosting web services.
• Azure Automation is for runbook automation, configuration management, and update management, not for hosting containerized models.
• Azure Logic Apps is a cloud service for building automated workflows that integrate apps and services; while it can trigger a deployed model, it is not the service hosting the model.

Microsoft Official Documentation: Azure Machine Learning. Model Deployment - Deploy a model to Azure Kubernetes Service. (Latest version) - Section: "When to use AKS."

Justification: Explicitly states that "Azure Kubernetes Service (AKS) provides a managed environment for deploying and scaling web services... Recommended for high-scale production deployments and when you need low-latency, auto-scaling, and GPU support."

Microsoft Official Documentation: Azure Kubernetes Service. Product Overview. (Latest version)

Justification: Confirms AKS is a managed Kubernetes offering for "deploying and managing containerized applications," which is the standard mechanism for hosting ML models for inference.

Microsoft Official Documentation: Azure Data Factory. What is Azure Data Factory? (Latest version)

Justification: Documents ADF's role as a "cloud-based ETL and data integration service" and not as a platform for hosting persistent web services or containerized models.

No: QnA Maker ingests data from specified sources, such as existing FAQ URLs, .pdf, .docx, or .tsv files, to create a knowledge base. It does not provide a mechanism to directly connect to and execute queries against a structured database like Azure SQL Database.

Yes: This describes the primary function of QnA Maker (now part of Azure AI Language's custom question answering). It is designed to map multiple phrasings of a question (similar questions) to a single, curated answer, ensuring a consistent response for all users.

No: In the context of Azure AI services, intent determination (classifying what a user wants to do, e.g., 'BookFlight') is the primary function of the Language Understanding (LUIS) service, now part of Conversational Language Understanding. QnA Maker's function is question matching—finding the best existing answer in a knowledge base for a user's question.

Microsoft. (2023). What is custom question answering? Azure AI services documentation. "Custom question answering provides cloud-based Natural Language Processing (NLP) capabilities that allow you to create a natural conversational layer over your data. It is used to find the most appropriate answer for any input from your custom knowledge base (KB) of information." and "Sources of data... can include existing content such as FAQ pages, product manuals, .tsv, .pdf, .docx files...".

Note: This documentation confirms the service's purpose (finding answers from a KB) and its supported data sources, which do not include direct queries to Azure SQL.

Microsoft. (2023). What is Conversational Language Understanding (CLU)? Azure AI services documentation. "Conversational Language Understanding (CLU) is a cloud-based conversational AI service that applies custom machine-learning intelligence to a user's conversational, natural language text to predict overall intent and extract important entities."

Note: This reference explicitly assigns the capability of "intent" prediction to LUIS/CLU, distinguishing it from QnA Maker's role.

Microsoft. (2023). LUIS and custom question answering. Azure AI services documentation. "Create a bot, using both LUIS and custom question answering. This type of bot can answer questions (custom question answering) and perform tasks (LUIS)."

Note: This document explicitly separates the services by their primary function: QnA Maker for answering questions and LUIS for performing tasks (which is based on intent determination).

Statement 1 (No): This is incorrect. The ability to include custom Python scripts as distinct components in a visual training pipeline is a primary feature of Azure Machine Learning designer (using the "Execute Python Script" module), not Automated ML.

Statement 2 (Yes): This is correct. Automated ML (AutoML) is specifically designed to automate the iterative, time-consuming tasks of model development. It provides a UI-driven, "no-code" interface that empowers users without deep programming or data science expertise to train and deploy high-performing models.

Statement 3 (No): This is incorrect. The capability to "visually connect datasets and modules on an interactive canvas" is the definition of Azure Machine Learning designer. AutoML uses a configuration-based UI but not a drag-and-drop canvas for pipeline construction.

Microsoft. (n.d.). What is automated machine learning (AutoML)? Azure Machine Learning Documentation. Retrieved October 24, 2025.

Section: "Overview"

Content (supports Statement 2): "Automated ML... empowers data scientists, analysts, and developers... to build ML models with high scale, efficiency, and productivity all while sustaining model quality... Automated ML also generates... models for users with or without a data science background."

Microsoft. (n.d.). What is Azure Machine Learning designer? Azure Machine Learning Documentation. Retrieved October 24, 2025.

Section: "Overview", Paragraph 1

Content (supports Statement 3 as False): "Azure Machine Learning designer provides a drag-and-drop canvas to build, test, and deploy machine learning models... The designer provides a visual canvas where you can drag and drop datasets and modules onto the canvas. You connect the modules to create a pipeline draft..."

Microsoft. (n.D.). Execute Python Script module reference. Azure Machine Learning designer. Retrieved October 24, 2025.

Section: "Overview"

Content (supports Statement 1 as False): "Use the Execute Python Script module in Azure Machine Learning designer to run custom Python code. With this module, you can write and run Python code that isn't currently available in the designer's built-in modules."

Chatbots can support voice input (Yes): Chatbots are not limited to text. When integrated with Speech-to-Text (STT) for input and Text-to-Speech (TTS) for output, they become "spoken dialogue systems" or "voicebots."

A separate chatbot is required for each communication channel (No): Modern conversational AI platforms are designed for omnichannel deployment. They use an architecture that decouples the core dialog logic and Natural Language Understanding (NLU) from the channel-specific adapters. This allows a single, centralized bot to be deployed simultaneously across multiple channels (e.t., web, SMS, Microsoft Teams, Slack) by formatting the input and output for each specific endpoint.

Chatbots manage conversation flows... (Yes): This describes a hybrid interaction model. Bots use NLU to interpret "natural language" (free text) from the user. They also use "constrained options" (such as buttons, quick replies, or menus) to guide the user, clarify intent, and manage the conversation flow efficiently.

Jurafsky, D., & Martin, J. H. (2024). Speech and Language Processing (3rd ed. draft). Stanford University.

Reference for Statement 1: Chapter 26, "Spoken Dialogue Systems." This chapter details the architecture of systems that "communicate with users via speech."

Reference for Statement 3: Chapter 27, "Task-Oriented Dialogue." Section 27.1 discusses dialogue system architectures, contrasting "natural language" interfaces with "command-based" (constrained) interfaces, noting that modern systems often combine these paradigms.

Ghandour, R., et al. (2021). "A Framework for the Development of Omnichannel Chatbots." 2021 IEEE 15th International Conference on Application of Information and Communication Technologies (AICT).

Reference for Statement 2: Section II.A, "Omnichannel Chatbot Architecture." The paper explicitly states: "The objective of the framework is to build the chatbot once and deploy it on all desired channels... It consists of a core bot and channel-specific adapters." (DOI: 10.1109/AICT52884.2021.9611985)

Adamopoulou, E., & Moussiades, L. (2020). "An Overview of Chatbot Technology." IFIP Advances in Information and Communication Technology, vol 584. Springer.

Reference for Statement 3: Section 4, "Chatbot Dialogue." This section reviews dialogue management techniques, including "menu-based" (constrained) and "NLP-based" (natural language) methods as primary interaction styles. (DOI: 10.1007/978-3-030-49186-4_30)

The Azure Machine Learning designer is a low-code/no-code tool within the Azure ML workspace. Its primary interface is a visual canvas where users build end-to-end ML pipelines by dragging, dropping, and connecting pre-built modules for tasks like data import, transformation, model training, and scoring.

• Option 3 (automatically selecting an algorithm...) is incorrect as this specifically describes Automated ML (AutoML), a separate feature.
• Option 4 (using a code-first notebook experience) is incorrect as this describes using Jupyter notebooks with the Azure ML SDK, which is the code-based alternative to the visual designer.

Microsoft. (2024, May 22). What is Azure Machine Learning designer? Microsoft Learn.

Section: Introduction, Paragraphs 1-2.

Quote: "Azure Machine Learning designer is a drag-and-drop interface... The designer provides a visual canvas where you can add datasets and modules... You connect the modules to create a pipeline draft..."

URL: https://learn.microsoft.com/en-us/azure/machine-learning/concept-designer

Microsoft. (2024, October 17). What is automated machine learning (AutoML)? Microsoft Learn.

Section: "How AutoML works."

Note: This official documentation confirms that "automatically selecting an algorithm" is the function of AutoML, distinguishing it from the designer.

URL: https://learn.microsoft.com/en-us/azure/machine-learning/concept-automated-ml

Duke University. (n.d.). AI for Everyone: Azure ML Designer. Duke AIPI (Artificial Intelligence for Product Innovation).

Section: "Azure ML Designer."

Quote: "Azure ML Designer... is a visual, drag-and-drop environment... In the Designer, you create an ML 'pipeline' on a visual canvas by dragging and dropping 'modules'..."

URL: https://aipi.pratt.duke.edu/azure-ml-designer

Microsoft. (2024, September 24). What are Jupyter notebooks in Azure Machine Learning? Microsoft Learn.

Section: Introduction.

Note: This documentation defines the "code-first notebook experience" as a separate component (Jupyter notebooks) within the Azure ML workspace, confirming it is distinct from the designer.

URL: https://learn.microsoft.com/en-us/azure/machine-learning/concept-notebooks

This question tests the fundamental differences between common computer vision tasks:

1. Image classification assigns one or more labels to an entire image. The scenario "separate images of polar bears and brown bears" requires classifying the whole image as either "polar bear" or "brown bear."
2. Object detection locates instances of objects within an image and draws a bounding box around them. This matches "determine the location of a bear," as it identifies where the bear is without detailing its exact shape.
3. Semantic segmentation is a pixel-level task. It classifies every pixel in an image as belonging to a specific class (e.g., "bear," "tree," or "sky"). This directly corresponds to "determine which pixels in an image are part of a bear."

Microsoft. (n.d.). Image analysis. Azure AI services documentation. Retrieved October 24, 2025.

Section: "Image classification" - "Classification assigns a class label to an entire image... for example, 'dog' or 'cat'."

Section: "Object detection" - "Object detection locates and identifies objects in an image. The model returns the coordinates of a bounding box for each object found."

Section: "Image segmentation" - "Segmentation creates a pixel-level mask for specific objects... in an image."

Stanford University. (n.d.). CS231n: Convolutional Neural Networks for Visual Recognition. Course Notes.

Section: "Course Notes: Image Classification" - "Image classification is the task of taking an input image and outputting a single label (or a distribution over labels) ... that describes the image."

Section: "Course Notes: Object Detection" - "This is the task of... localizing all objects (giving their bounding boxes)."

Section: "Course Notes: Semantic Segmentation" - "Semantic Segmentation is the task of labeling every pixel in the image with a category label."

Garcia-Garcia, A., Orts-Escolano, S., Oprea, S., Villena-Martinez, V., & Garcia-Rodriguez, J. (2017). A review on semantic segmentation using deep learning. Electronics, 6(4), 75.

DOI: https://doi.org/10.3390/electronics6040075

Section: 1. Introduction, Paragraph 1. "Semantic segmentation, or fully pixel-level classification, is the task of assigning a semantic label (e.g., ‘person’, ‘car’, ‘dog’) to every pixel in an image."

The tasks align with core LLM capabilities. Classifying involves assigning a predetermined category, such as a genre, to an input text. Summarizing is the process of extracting the most important information and presenting it concisely, often using a format like bullet points. Generating refers to creating new, novel, or creative content that goes beyond simple summarization or classification, such as developing advertising slogans from a product description. Translating is not used as none of the tasks involve converting text from one human language to another.

Reference for Classifying:

Source: Speech and Language Processing (3rd ed. draft), Jurafsky, D., & Martin, J. H. (2023). Section: Chapter 5: Text Classification and Sequence Labeling.

Details: This chapter covers the application of models (including LLMs) to text classification tasks, where the model determines the category (like genre) of the input text.

Reference for Summarizing:

Source: OpenAI Documentation: Azure OpenAI Service (Microsoft Official Documentation). Section: Core Concepts: Summarization.

Details: Official vendor documentation often categorizes generating concise bullet points from a long text as a form of extractive or abstractive summarization task.

Reference for Generating:

Source: Attention Is All You Need, Vaswani, A., et al. (2017). arXiv:1706.03762. DOI: https://doi.org/10.48550/arXiv.1706.03762.

Details: This foundational paper on the Transformer architecture (the basis for LLMs) describes the model's ability to produce new sequences (text generation), which encompasses creative tasks like slogan creation.

Reference for Classifying, Generating, and Summarizing as LLM tasks:

Source: Stanford CS224N: Natural Language Processing with Deep Learning (Stanford University Courseware). Topic: Large Language Models and their applications.

Details: University course materials routinely delineate these three operations—Classification (e.g., sentiment, topic), Summarization (e.g., abstractive, extractive), and Generation (e.g., creative writing, dialogue)—as fundamental applications of modern LLMs.

The solution correctly maps the three fundamental types of machine learning models to their definitions.

• Regression is a supervised learning task used to predict a continuous numeric value, such as the price of a house or a future temperature.
• Classification is also a supervised learning task, but it is used to predict a discrete category or class, such as whether an email is 'spam' or 'not spam', or if a tumor is 'benign' or 'malignant'.
• Clustering is an unsupervised learning task. It does not use pre-defined labels; instead, it analyzes the input data to identify natural groupings (clusters) of items based on their shared features or similarities.

Microsoft. (2024). "What is machine learning?" Azure Machine Learning Documentation. Microsoft.

Section: "Machine learning model types" > "Supervised learning"

Quote/Paraphrase: This documentation explicitly defines regression as a supervised method for predicting continuous values (e.g., price, sales). It defines classification as a supervised method for predicting categories (e.g., yes/no, true/false). It defines clustering as an unsupervised method used to discover structure and group items into clusters based on similarity.

James, G., Witten, D., Hastie, T., & Tibshirani, R. (2021). An Introduction to Statistical Learning: with Applications in R (Second Edition). Springer.

Chapter 2, Section 2.1.2 "Supervised and Unsupervised Learning": This section distinguishes the two main types. It states that supervised learning involves building a model for predicting an output based on inputs, further breaking this down into regression problems (predicting a quantitative or numeric output) and classification problems (predicting a qualitative or categorical output).

Chapter 12, Section 12.1 "Unsupervised Learning": This section defines unsupervised learning as a setting with only feature measurements (X) and no response variable (Y). The goal is described as finding interesting patterns or groups, which directly relates to clustering.

Ng, A. (2023). "Course Notes: CS229 - Machine Learning." Stanford University.

Section: "Part I: Supervised Learning"

Quote/Paraphrase: The notes define supervised learning as the task of learning a function that maps inputs to outputs given a set of input-output pairs. It specifies that if the target output (label) is continuous (e.g., price), the task is regression. If the target output is discrete (e.g., 'cat' or 'dog'), the task is classification.

Section: "Part IX: Unsupervised Learning"

Quote/Paraphrase: This section describes unsupervised learning as the process of finding structure in unlabeled data. The canonical example provided is clustering, such as grouping news articles by topic or customers by preferences.

Azure AI Vision is the correct service for Classify and label images as it provides image analysis, tagging, object detection, and content description (Source 3.1, 3.2). Azure OpenAI Service is the platform for models like GPT, which are fundamentally used to create conversational responses and generate text (Source 1.2, 1.4). Azure AI Speech is dedicated to speech-to-text and text-to-speech capabilities, including real-time transcription of spoken audio (Source 2.1, 2.2).

Microsoft Learn. "What is Azure OpenAI in Azure AI Foundry Models?". Section: Features overview and Prompts & completions. (Source 1.2 in search results).

Microsoft Learn. "What is the Speech service? - Azure AI services". Paragraph 1. (Source 2.1 in search results).

Microsoft Learn. "What is the Speech service? - Azure AI services". Section: Speech to text. (Source 2.2 in search results).

Microsoft Docs. "What is Azure AI Vision? - Azure AI services". Section: Image Analysis. (Source 3.2 in search results).

ITMAGINATION. "Azure AI Vision: Technology Overview, Use Cases, and Pricing". Section: What It Does. (Source 3.1 in search results).

The sequence represents the standard, fundamental workflow for creating a machine learning model before it can be deployed.

1. Data Preparation: Raw data must first be collected, cleaned, transformed, and structured (e.g., feature engineering, splitting into training and test sets). A model cannot be trained on unprepared data.
2. Model Training: The prepared training dataset is used to "teach" an algorithm, allowing it to learn patterns and create the model.
3. Model Evaluation: After training, the model's performance must be validated using a separate test dataset (which it has not seen before) to measure its accuracy and effectiveness. A model is only deployed if its evaluation results meet the required business objectives.
4. The other options, model retraining and data encryption, are incorrect for this sequence. Model retraining is a post-deployment process performed to update a model when its performance degrades over time (model drift). Data encryption is a security process applied to data at rest or in transit, not a sequential step in the model creation pipeline itself.

Source: Microsoft (n.d.). What is Azure Machine Learning designer? Microsoft Docs.

Reference: In the section titled "Build a pipeline," the documentation describes the visual workflow: "Pipelines start with your data... you can connect these datasets to other modules which prepare data, train models, and evaluate models." This officially confirms the sequential order of these three core processes.

Source: Amershi, S., Begel, A., Bird, C., et al. (2019). Software Engineering for Machine Learning: A Case Study. 2019 IEEE/ACM 41st International Conference on Software Engineering: Software Engineering in Practice (ICSE-SEIP).

Reference: Page 292 (Section III.A, "AI Development Workflow"). The paper describes the typical ML workflow stages identified through case studies, stating: "...the AI development workflow includes data preparation (e.g., collect, clean, label), model training (e.g., feature engineering, training), and model evaluation..."

DOI: https://doi.org/10.1109/ICSE-SEIP.2019.00018

Source: Microsoft (n.d.). The machine learning lifecycle. Microsoft Docs.

Reference: The document explicitly outlines the stages of the machine learning lifecycle. The core model creation process is described in this order: 3. Data preparation, 4. Model training, and 5. Model evaluation, which all occur before stage 6. Model deployment. This clearly separates the pre-deployment steps from post-deployment activities like "Model monitoring" (which leads to retraining).

Transparency means that the system's operation, particularly the factors that influence its output, should be understandable, allowing for inspection and explanation. Recording the decision-making process to identify the reasoning behind a quote directly aligns with this principle.

Privacy and security focuses on ensuring customer data is protected and that access is appropriately controlled. Restricting visibility of a customer's personal information to only relevant staff is a core component of both data privacy and system security.

Inclusiveness requires that AI systems should empower people and engage them fairly, including addressing accessibility needs for people with disabilities. Ensuring the system works with screen readers and other assistive technology is a direct application of the inclusiveness principle.

Source: Microsoft. Microsoft Responsible AI Standard, v2. (Section 2.2: Transparency, Section 2.5: Privacy and Security, Section 2.3: Inclusiveness).

Data:

Transparency: "AI systems should be understandable, and their operation should be open to inspection." Requirement for recording decision-making logic aligns with the need for interpretability and explainability.

Privacy and security: "AI systems should protect customer data against unauthorized access and disclosure." Restricting access to personal information aligns directly with the goal of data protection.

Inclusiveness: "AI systems should empower everyone, engage people fairly, and address accessibility needs for people with disabilities." System accessibility for screen readers and assistive technology is a specific requirement under accessibility.

Source: IEEE. P7003 Standard for Algorithmic Bias Considerations. (Clause 5.4, 6.1).

Data: The standard discusses accessibility as part of a comprehensive view of fairness and ethical design, supporting the link between using assistive technologies (screen readers) and the Inclusiveness principle, which aims to avoid exclusion.

Source: The Alan Turing Institute. Understanding Transparency in AI: A Guide to Policy and Practice. (Section 3.1, p. 12).

Data: Defines transparency as the degree to which an AI system's inner workings are understandable, which necessitates explainability and interpretability—the requirement to record the decision-making process is a fundamental step toward achieving this.

Image classification assigns a single label to an entire image, which is the foundational step for tasks like generating captions by identifying the main content. Optical Character Recognition (OCR) is specifically designed to recognize and extract text from images, making it the appropriate choice for retrieving movie titles from posters. Object detection involves finding the location (usually with bounding boxes) of one or more specific objects within an image and labeling them, which directly corresponds to the task of locating vehicles.

"Deep Learning" by I. Goodfellow, Y. Bengio, and A. Courville (MIT Press, 2016). Section 9.1, "The Classification Task," describes image classification as mapping an image to a single category label. Section 9.2, "Object Detection," introduces bounding boxes for localization.

"Computer Vision: Algorithms and Applications" by R. Szeliski (Springer, 2011). Chapter 14, "Recognition," discusses object recognition and localization methods. Pages 771-773 cover the distinction between image classification and object detection with localization.

"Fundamentals of Computer Vision" (MIT OpenCourseWare 6.819/6.869). Lecture Notes on Vision and Language/Image Captioning state that image classification and object recognition provide the input (concepts, objects) necessary for higher-level tasks like caption generation.

ISO/IEC 2382-37:2017 Information technology — Vocabulary — Part 37: Biometrics. Definition 37.1.3.1.2 defines OCR as the ability to interpret characters on a document or image.

"Optical Character Recognition: Technology and Applications" in Machine Vision and Applications (Springer, 2019, Special Issue). Highlights OCR's core function as the conversion of scanned images of text into machine-encoded text (extraction of titles from posters is a direct application). (e.g., DOI: 10.1007/s00138-019-01031-4).

The actions correspond directly to the core capabilities of the respective Azure AI Services.

• Azure AI Speech is responsible for speech-to-text (and text-to-speech), making it the correct service for "Convert spoken requests into text" (Source 1.1, 1.4).
• Azure AI Language (specifically its Conversational Language Understanding or CLU component, which succeeded LUIS) is designed for Natural Language Processing (NLP) tasks (Source 4.2). This includes recognizing the overall intent (e.g., "book a flight") from a user's utterance (the input sentence) and extracting key data points called entities (e.g., "Seattle" as a location) to fulfill that request (Source 2.1, 2.2, 2.4). Therefore, both "Identify the intent..." and "Apply intent to entities and utterances" map to Azure AI Language.
• Azure AI Translator is for machine translation, which is not listed in the actions.

1.1, 1.4, 4.1 (Official vendor documentation - Microsoft Learn): Confirms that Azure AI Speech includes the Speech to text feature, which converts spoken audio into text.

2.1, 2.2, 2.4, 4.2, 4.3 (Official vendor documentation - Microsoft Learn): Verifies that Azure AI Language (via CLU/LUIS) is used for Natural Language Processing (NLP) tasks. Specifically, it states that an intent is the action the user wants to perform, an utterance is the user's input, and entities are the relevant data points, all of which are managed by the Language service to predict and apply intent.

3.1, 3.2 (Official vendor documentation - Microsoft Learn): Confirms that Azure AI Translator is a neural machine translation service for translating text and documents, which does not match any of the given actions.

Fairness is the principle that AI systems should treat all people equitably, avoiding any bias or discrimination based on gender, race, or other protected characteristics.

Privacy and security is the principle that AI systems must be secure and respect user privacy. This includes protecting personal data from exposure and ensuring it is only accessible to authorized individuals.

Transparency is the principle that AI systems should be understandable. This involves enabling users to know why a system made a particular decision, often through logging and the ability to audit the decision-making process.

Microsoft. (n.d.). The six principles of responsible AI. Microsoft Responsible AI.

Reference for Fairness: In the "Fairness" section, it is stated that this principle addresses "providing equal treatment to all people, and avoiding bias based on... gender, race, or other factors..."

Reference for Privacy and security: In the "Privacy and security" section, the principle is defined by the need for systems to "protect data and honor privacy... ensuring that personal data isn't exposed."

Reference for Transparency: In the "Transparency" section, the principle is explained as helping "people understand how [AI systems] work and why they make the decisions they do." This includes "awareness of how data was used... and a willingness to open the system to scrutiny."

The solution correctly maps each machine learning model type to its corresponding task:

1. Clustering: This is an unsupervised learning model used to find natural groupings or "clusters" in data. The task "Assign categories to passengers" does not have a predefined correct output; instead, the goal is to discover segments based on similarity, which is the definition of clustering.
2. Regression: This is a supervised learning model used to predict a continuous numerical value. The task "Predict the amount of consumed fuel" requires predicting a specific number (e.g., 5,000 liters), making it a regression problem.
3. Classification: This is a supervised learning model used to predict a discrete category or class. The task "Predict whether a passenger will miss their flight" requires a categorical prediction (e.g., "Yes" or "No"), which is a classic binary classification problem.

Microsoft. (n.d.). Machine learning basics: What is regression?. Microsoft Learn (AI-900 path). This document states, "Regression is a form of supervised machine learning in which you train a model to predict a numeric value." This directly supports matching Regression to "Predict the amount of consumed fuel."

Microsoft. (n.d.). Machine learning basics: What is classification?. Microsoft Learn (AI-900 path). This official documentation notes, "Classification is a form of supervised machine learning in which you train a model to predict which category (or class) an item belongs to." This supports matching Classification to "Predict whether..." (a yes/no category).

Microsoft. (n.d.). Machine learning basics: What is clustering?. Microsoft Learn (AI-900 path). This document defines clustering as an "unsupervised machine learning in which you train a model to group similar items into clusters." This supports matching Clustering to "Assign categories to passengers," as the model creates the categories.

Ng, A. (n.d.). CS229 Lecture Notes 1. Stanford University.

Part I ("Supervised Learning"): Distinguishes between regression (predicting a continuous-valued output) and classification (predicting a discrete-valued output). This reference supports the separation of the regression (amount of fuel) and classification (will miss flight) tasks.

Part V ("Unsupervised Learning"): Describes clustering as a problem where "we are given a data set and we’d like to automatically find structure in the data," such as grouping it into clusters. This supports the "Assign categories" task.

Hastie, T., Tibshirani, R., & Friedman, J. (2009). The Elements of Statistical Learning: Data Mining, Inference, and Prediction (2nd ed.). Springer.

Chapter 2, Section 2.3 ("Supervised and Unsupervised Learning"): This foundational text formally distinguishes the goals. Supervised learning (classification, regression) aims to predict a specific, known output or response. Unsupervised learning (clustering) has "no outcome variable, but rather... to describe the associations and patterns among a set of input variables." This confirms the fundamental logic used in the answer.

The Azure AI Speech service provides core functions like Speech to text (transcribing spoken audio into written text) and Text to speech (converting input text into natural-sounding, synthesized speech, which provides the spoken response). The Azure AI Language service includes features like Conversational Language Understanding (CLU), which is specifically designed to build natural language understanding models to predict the overall intention of an incoming utterance (identify a user's intent) from text input. Azure AI Translator Text is used for text-to-text translation and is not required for the core tasks of converting speech to text, identifying intent, or generating a spoken response.

Azure AI Speech Capabilities:

Microsoft Learn. "What is the Speech service? - Azure AI services." Snippet 1.1 states the service can "transcribe speech to text with high accuracy" and "produce natural-sounding text to speech voices."

Microsoft Learn. "Speech service documentation - Tutorials, API Reference." Snippet 1.3 explicitly lists "Speech to text" and "Text to speech" as features.

Azure AI Language Service Capabilities (Intent Identification):

Microsoft Learn. "What is Azure AI Language - Azure AI services." Snippet 2.1 specifies that Conversational language understanding (CLU), a feature of the Language service, "enables users to build custom natural language understanding models to predict the overall intention of an incoming utterance and extract important information from it."