Free Microsoft AI-900 Practice Exam with Questions & Answers | Set: 5

According to the Microsoft Azure AI Fundamentals (AI-900) syllabus and Microsoft Learn module “Describe features of natural language processing (NLP) workloads on Azure,” translation is a core NLP workload that converts text from one language into another while maintaining meaning and context.

In this scenario, the website stores reviews in English and must present them in the user’s native language based on geographical location. This directly requires a translation workload, which uses Azure Cognitive Services — specifically, the Translator service — to automatically translate content dynamically for each user.

Other options explained:

B. Language modeling involves predicting the next word in a sentence or understanding linguistic patterns; it’s used in model training, not translation.

C. Key phrase extraction identifies main ideas in text, not language conversion.

D. Speech recognition converts spoken words into written text but does not perform translation or handle geographic adaptation.

Microsoft’s Translator service supports real-time text translation, multi-language detection, and context preservation, making it ideal for global websites. The AI-900 study guide emphasizes translation as one of the most common NLP workloads, enabling applications to break language barriers and enhance accessibility for diverse audiences.

Therefore, based on official Microsoft Learn material, the correct answer is:

✅ A. translation.

Safety system.

According to the Microsoft Learn documentation and the AI-900: Microsoft Azure AI Fundamentals official study guide, the safety system layer in generative AI architecture plays a crucial role in monitoring, filtering, and mitigating harmful or unsafe model outputs. This layer works alongside the model and user experience layers to ensure that generative AI systems—such as those powered by Azure OpenAI—produce responses that are safe, aligned, and responsible.

The safety system layer uses various techniques including content filtering, prompt moderation, and policy enforcement to prevent outputs that could be harmful, biased, misleading, or inappropriate. It evaluates both user inputs (prompts) and model-generated outputs to identify and block unsafe or unethical content. The system might use predefined rules, classifiers, or human feedback signals to decide whether to allow, modify, or stop a response.

In contrast, the other layers serve different purposes:

The model layer contains the core large language or generative model (e.g., GPT or DALL-E) that processes inputs and produces outputs.

The metaprompt and grounding layer ensures the model’s responses are contextually relevant and factually supported, often linking to organizational data sources or system prompts.

The user experience layer defines how users interact with the AI system, including the interface and conversational flow, but does not manage safety enforcement.

Therefore, the layer that uses system inputs and context to mitigate harmful outputs from a generative AI model is the Safety system layer.

This aligns with Microsoft’s responsible AI principles—Fairness, Reliability and Safety, Privacy and Security, Inclusiveness, Transparency, and Accountability—ensuring generative AI operates ethically and safely.

Confidence.

According to the Microsoft Azure AI Fundamentals (AI-900) Official Study Guide and the Microsoft Learn module “Explore computer vision in Microsoft Azure,” the confidence score represents the calculated probability that a model’s prediction is correct. In image classification, when an AI model analyzes an image and assigns it to a specific category, it also produces a confidence value—a numerical probability (usually between 0 and 1) indicating how certain the model is about its prediction.

For example, if an image classification model identifies an image as a “cat” with a confidence of 0.92, it means the model is 92% certain that the image depicts a cat. The confidence value helps developers and users understand the model’s certainty level about its classification output.

Microsoft Learn emphasizes that in Azure Cognitive Services—such as the Custom Vision Service—each prediction result includes both the predicted label (class) and a confidence score. These confidence scores are essential for evaluating model performance and determining thresholds for automated decisions (e.g., accepting predictions only above a 0.8 probability).

Let’s evaluate the other options:

Accuracy: This is an overall performance metric measuring the percentage of correct predictions across the dataset, not a probability for a single prediction.

Root Mean Square Error (RMSE): This is a metric for regression models, not classification tasks. It measures average error magnitude between predicted and actual values.

Sentiment: This is a type of prediction (positive, negative, neutral) in text analysis, not a probability metric.

Therefore, based on Microsoft’s AI-900 training materials and Azure Cognitive Services documentation, the calculated probability of a correct image classification is called Confidence, which expresses how sure the model is about its prediction for a specific input.

In Azure OpenAI Service, the temperature parameter directly controls the creativity and determinism of responses generated by models such as GPT-3.5. According to the Microsoft Learn documentation for Azure OpenAI models, temperature is a numeric value (typically between 0.0 and 2.0) that determines how “random” or “deterministic” the output should be.

A lower temperature value (for example, 0 or 0.2) makes the model’s responses more deterministic, meaning the same prompt consistently produces nearly identical outputs.

A higher temperature value (for example, 0.8 or 1.0) encourages creativity and variety, causing the model to generate different phrasing or interpretations each time it responds.

When a question specifies the need for more deterministic responses, Microsoft’s guidance is to decrease the temperature parameter. This adjustment makes the model focus on the most probable tokens (words) rather than exploring less likely options, improving reliability and consistency—ideal for business or technical applications where consistent answers are essential.

The other parameters serve different purposes:

Frequency penalty reduces repetition of the same phrases but does not control randomness.

Max response (max tokens) limits the maximum length of the generated output.

Stop sequence defines specific tokens that tell the model when to stop generating text.

Thus, the correct and Microsoft-verified completion is:

“You can modify the Temperature parameter to produce more deterministic responses from a chat solution that uses the Azure OpenAI GPT-3.5 model.”

To reduce operator load with a chatbot for predefined answers:

QnA Maker provides the knowledge base for answering questions automatically.

Azure Bot Service hosts and manages the chatbot interface to interact with users.Text Analytics and Translator Text are not required for basic Q & A chatbots.

In Azure Machine Learning designer, a low-code drag-and-drop interface, users can visually build machine learning workflows. According to the AI-900 study guide and Microsoft Learn module “Create and publish models with Azure Machine Learning designer”, two key components that can be dragged onto the designer canvas are datasets and modules.

Datasets (A): These are collections of data that serve as the input for training or evaluating models. They can be registered in the workspace and then dragged onto the canvas for use in transformations or model training.

Modules (D): These are prebuilt processing and modeling components that perform operations such as data cleaning, feature engineering, model training, and evaluation. Examples include “Split Data,” “Train Model,” and “Evaluate Model.”

Compute (B) and Pipeline (C) are not drag-and-drop items within the designer. Compute targets are infrastructure resources used to run the pipeline, while a pipeline represents the overall workflow, not a component that can be added like a dataset or module.

Hence, the correct answers are A. Dataset and D. Module.

The described app identifies the coordinates of a product within an image of a supermarket shelf. This scenario directly corresponds to the object detection capability of Azure AI Custom Vision. As per the Microsoft Learn module “Train a Custom Vision model”, the object detection project type allows developers to train models that can both detect and locate objects within an image. It returns bounding box coordinates along with predicted labels for each detected item.

In this use case, the app doesn’t just classify what products are present—it needs the position of the product (coordinates). That function distinguishes object detection from classification. Classification simply assigns a label to the entire image, while object detection provides spatial information for multiple items in one image.

The other options are incorrect:

B. Azure AI Computer Vision Read and C. Azure AI Computer Vision OCR are used for extracting text from images, not locating objects.

D. Azure AI Custom Vision classification only categorizes images but cannot determine where objects appear.

Therefore, to build an app that finds and locates products in images, the correct choice is A. Azure AI Custom Vision object detection.

Microsoft’s Responsible AI principles are the foundation for developing and deploying ethical and trustworthy AI systems. The six key principles are Fairness, Reliability and Safety, Privacy and Security, Inclusiveness, Transparency, and Accountability. Each principle guides specific practices for ensuring AI systems operate responsibly in real-world applications like automated insurance quoting systems.

Transparency – This principle ensures that the AI’s decisions can be understood and explained. Recording the decision-making process and enabling staff to trace how a quote was generated aligns with transparency. It allows stakeholders to interpret the reasoning behind model outputs, ensuring that the AI behaves predictably and ethically.

Privacy and Security – This principle focuses on protecting personal data and ensuring that sensitive information is handled responsibly. Limiting access to customer data only to authorized personnel maintains compliance with privacy laws (like GDPR) and safeguards against misuse. Microsoft emphasizes that AI systems should maintain strict control over data visibility and integrity.

Inclusiveness – This principle ensures that AI systems are accessible to all users, including people with disabilities. By supporting screen readers and assistive technologies, the system ensures equal access to information and services for every customer. Inclusiveness prevents discrimination and promotes accessibility, both of which are central to Microsoft’s Responsible AI strategy.

Thus, the correct mapping of principles is:

Decision process → Transparency

Personal information visibility → Privacy and Security

Accessibility via screen readers → Inclusiveness.

According to the Microsoft Azure AI Fundamentals (AI-900) curriculum and Azure OpenAI documentation, the appropriate service for detecting and managing harmful, unsafe, or inappropriate content in text, images, or other generative AI outputs is Azure AI Content Safety.

Azure AI Content Safety is designed to automatically detect potentially harmful material such as hate speech, violence, self-harm, sexual content, or profanity. It ensures that generative AI applications like chatbots, image generators, and content creation tools comply with Microsoft’s Responsible AI principles — specifically Reliability & Safety and Accountability.

This service integrates directly with the Azure OpenAI Service, meaning that when developers build AI solutions using models like GPT-4 or DALL·E, they can use Content Safety to filter and moderate both input prompts and model outputs. This protects users from unsafe or offensive content generation.

Let’s analyze why the other options are incorrect:

A. Face – The Face service detects and analyzes human faces in images or videos. It is unrelated to moderating harmful textual or generative content.

B. Video Analysis – This service analyzes video streams to detect objects, actions, or events but not inappropriate or harmful text or imagery from AI models.

C. Language – The Azure AI Language service focuses on text understanding tasks like sentiment analysis, entity recognition, and translation, not content safety filtering.

Therefore, per Microsoft Learn’s official AI-900 guidance, when identifying or filtering harmful content in a generative AI solution built with Azure OpenAI, the correct and verified service to use is Azure AI Content Safety.

QnA Maker is an Azure Cognitive Service used to build question-and-answer knowledge bases from structured and unstructured documents, such as FAQs, product manuals, or webpages. According to the AI-900 study guide and Microsoft Learn module “Build a knowledge base with QnA Maker”, this service allows you to extract question-answer pairs from existing data sources like FAQ pages, PDF files, or Word documents.

In this scenario, you have:

A product troubleshooting guide (Word document)

A FAQ webpage

QnA Maker can automatically read both sources, extract relevant Q & A pairs, and create a knowledge base that your chatbot can use to respond to user queries intelligently.

To clarify the other options:

A. Azure Bot Service provides the chatbot interface and conversation logic but doesn’t extract knowledge from documents.

B. Language Understanding (LUIS) identifies intents and entities in natural language input, but it’s not used to read document content.

C. Text Analytics is used for key phrase extraction and sentiment analysis, not Q & A creation.

Therefore, the correct service for processing FAQ-style and document-based content into a question-answering bot is QnA Maker.