Microsoft Azure AI Engineer Associate AI-102 AI-102 Questions 451–525 | Page 7/11

451

MCQhard

You are using Azure AI Language's custom question answering feature to build a FAQ bot. The knowledge base contains many question-and-answer pairs. Users sometimes ask questions that are paraphrases of the stored questions. You need to improve the likelihood that the bot returns the correct answer for paraphrased questions. Which action should you take?

A.Enable active learning and accept all suggested question variants.

B.Add alternate questions to each question-and-answer pair in the knowledge base.

C.Switch the knowledge base to use a different language model for embeddings.

D.Increase the confidence threshold for answer returns to 90.

AnswerB

Alternate questions allow you to provide multiple phrasings for the same answer. When a user asks a paraphrased question, the service can match it to an alternate question and return the associated answer. This directly improves recall for variations in wording without changing the underlying model or adding unrelated content.

Why this answer

Adding alternate questions to each question-and-answer pair is the most direct way to improve matching for paraphrased queries. The service uses these variants to better understand different ways users might ask the same thing. Raising the threshold, accepting all active learning suggestions, or trying to change the embedding model do not reliably address the need to match paraphrases.

Exam trap

The trap here is thinking that raising the confidence threshold improves answer quality for paraphrases, when it actually makes the bot stricter and more likely to return no answer.

Full explanation →

452

MCQeasy

You are building an agent with the Azure AI Agents SDK that must support multi-turn conversations for a help desk scenario. The agent should remember details a user provided earlier in the same conversation, such as their device model. You need to manage conversation state. What should you do?

A.Include the entire conversation history in every request and omit thread creation.

B.Enable a file search tool so the agent can look up the device model from uploaded files.

C.Store the device model in the agent's instructions each time the user mentions it.

D.Create a thread for the conversation and add messages to it as the user and agent exchange turns.

AnswerD

Threads persist the message history for a conversation, so the agent can reference earlier user details such as a device model when answering later questions. Each new user message is added to the same thread, and the agent's responses are stored there as well. This is the standard mechanism for maintaining multi-turn context in the Azure AI Agents SDK.

Why this answer

Threads are the Azure AI Agents SDK mechanism for multi-turn state. By creating a thread and adding each message to it, the agent has access to prior turns and can recall details like a device model. The other approaches either bypass platform state management, misuse instructions, or apply a retrieval tool to a problem that requires conversation history.

Exam trap

The trap here is confusing document retrieval with conversational memory, when only thread-managed history preserves earlier user statements.

Full explanation →

453

Multi-Selectmedium

Which TWO Azure services can be used to implement a conversational AI solution that understands user intent and responds appropriately?

Select 2 answers

A.Azure Bot Service

B.Conversational Language Understanding

C.Azure AI Speech-to-Text

D.Azure AI Translator

E.Azure AI Search

AnswersA, B

Azure Bot Service provides the conversational orchestration layer, hosting the bot and connecting channels to language models. It satisfies the requirement to respond appropriately by routing user messages to intent recognition and returning generated replies.

Why this answer

Azure Bot Service (A) is correct because it provides the bot framework and channel integration needed to host a conversational AI solution that receives user messages and returns appropriate responses across channels like Teams, web chat, and Slack. Conversational Language Understanding (B), part of Azure AI Language, is correct because it is specifically designed to extract user intent and entities from natural language utterances, which is the core capability required for understanding what the user wants. Together, these services directly address intent recognition and conversational response handling.

Azure AI Speech-to-Text (C) only transcribes spoken audio into text and does not determine intent or generate conversational responses. Azure AI Translator (D) performs language translation and does not provide intent understanding or dialogue management. Azure AI Search (E) is a search indexing and retrieval service, not a conversational intent or bot response service.

Exam trap

The trap here is that candidates often confuse Azure AI Speech-to-Text (a transcription service) with a conversational AI solution, but it lacks intent recognition and response generation capabilities.

Full explanation →

454

Multi-Selecthard

Which THREE components are required to build a custom copilot using Microsoft Copilot Studio that can answer questions from a SharePoint document library?

Select 3 answers

A.A Microsoft Copilot Studio copilot

B.A knowledge source configured in Copilot Studio (e.g., Azure Cognitive Search)

C.A Power Automate flow to trigger the copilot

D.A SharePoint site with the documents

E.An Azure OpenAI Service deployment

AnswersA, B, D

The copilot is the conversational interface.

Why this answer

A Microsoft Copilot Studio copilot is the core conversational interface that orchestrates the interaction. Without the copilot itself, there is no runtime environment to process user queries, manage conversation state, or invoke configured knowledge sources. It is the mandatory host for all custom copilot functionality.

Exam trap

A common misconception is that you must provision an Azure OpenAI Service deployment to use generative AI in Copilot Studio, when in fact the platform provides its own managed GPT models and only requires a knowledge source like SharePoint.

Full explanation →

455

MCQhard

A company uses Azure OpenAI to generate product descriptions. They notice that the model occasionally produces descriptions that include false claims about product features. The company needs to reduce the frequency of these inaccuracies without changing the training data. Which parameter adjustment would be most effective?

A.Increase the top_p parameter

B.Increase the max_tokens parameter

C.Decrease the temperature parameter

D.Increase the frequency_penalty parameter

AnswerC

Lowering temperature reduces sampling randomness, so the model favours higher-probability tokens and produces more grounded, less speculative text. This directly addresses the false product claims without altering training data, satisfying the stem's constraint of no training-data changes.

Why this answer

Decreasing the temperature parameter reduces the randomness of the model's output, making it more deterministic and less likely to generate creative but factually incorrect statements. This directly addresses the need to reduce false claims without modifying training data, as lower temperature forces the model to rely on its most probable (and typically more accurate) token predictions.

Exam trap

The trap here is that candidates often confuse temperature with creativity or length control, assuming that increasing randomness (higher temperature) or extending output length (max_tokens) will somehow improve accuracy, when in fact lower temperature is the standard parameter for reducing hallucinations.

How to eliminate wrong answers

Option A is wrong because increasing top_p (nucleus sampling) expands the set of candidate tokens considered, which increases output diversity and can actually worsen factual inaccuracies by allowing less probable tokens. Option B is wrong because increasing max_tokens only extends the maximum length of the generated text; it does not influence the factual accuracy or creativity of the content. Option D is wrong because increasing frequency_penalty penalizes tokens that have already appeared, reducing repetition but not addressing the root cause of hallucinated or false claims.

Full explanation →

456

MCQeasy

You are developing a solution that uses Azure AI Language to detect the language of incoming text messages. The messages may contain mixed languages within a single document. You need to ensure the API returns the detected language and a confidence score for each document. Which request should you make?

A.POST to the analyze-text endpoint with kind set to EntityRecognition.

B.POST to the analyze-text endpoint with kind set to KeyPhraseExtraction.

C.POST to the analyze-text endpoint with kind set to SentimentAnalysis.

D.POST to the analyze-text endpoint with kind set to LanguageDetection.

AnswerD

The LanguageDetection task in the analyze-text endpoint returns the detected language and a confidence score for each document. This is the correct way to invoke language detection in the unified Language service. It handles mixed-language documents by returning the dominant language and its confidence score, which meets the requirement.

Why this answer

Language detection is performed by setting the kind parameter to LanguageDetection in the analyze-text request. This returns the detected language and a confidence score for each document. Other tasks like EntityRecognition, KeyPhraseExtraction, and SentimentAnalysis serve different purposes and do not provide language detection.

Exam trap

The trap here is assuming that any analyze-text task can detect language, when only the LanguageDetection kind returns the language and confidence score.

Full explanation →

457

MCQmedium

A company is implementing a question-answering system using Azure AI Language Service. They have a set of FAQ documents in PDF format. Which feature should they use to automatically generate question-answer pairs?

A.Key Phrase Extraction

B.Extractive Summarization

C.Custom Question Answering

D.Conversational Language Understanding

AnswerC

Custom Question Answering ingests source documents such as PDFs and automatically generates question-answer pairs from their content, optionally with follow-up prompts. This satisfies the stem's requirement to build a knowledge base from FAQ documents without manually authoring every pair.

Why this answer

Custom Question Answering (C) is the correct feature because it is specifically designed to ingest semi-structured content like FAQ PDFs and automatically generate question-answer pairs. It uses a built-in extraction pipeline that parses the document structure (e.g., headings, bullet points) to identify likely questions and their corresponding answers, which can then be reviewed and refined in the Azure Language Studio portal.

Exam trap

The trap here is that candidates confuse Custom Question Answering with Conversational Language Understanding (CLU), but CLU is for intent classification and entity extraction in dialog flows, not for automatic QnA pair generation from documents.

How to eliminate wrong answers

Option A is wrong because Key Phrase Extraction identifies important terms or concepts in text but does not generate question-answer pairs; it returns a list of key phrases without any relational mapping. Option B is wrong because Extractive Summarization produces a condensed version of the original text by selecting salient sentences, not by creating question-answer pairs from FAQ documents. Option D is wrong because Conversational Language Understanding (CLU) is designed to interpret user intents and extract entities from natural language utterances in a conversational flow, not to automatically generate question-answer pairs from static documents.

Full explanation →

458

MCQhard

You are building a generative AI application that must process large volumes of PDF documents and generate summaries using Azure OpenAI. The solution must be cost-effective and handle variable workloads. Which architecture should you recommend?

A.Use Azure Kubernetes Service (AKS) with a persistent node pool of GPU nodes.

B.Use Azure Functions with a consumption plan to trigger processing jobs and call Azure OpenAI.

C.Deploy a GPU-enabled virtual machine and run the summarization jobs sequentially.

D.Use Azure Logic Apps to iterate through documents and call Azure OpenAI.

AnswerB

Azure Functions on a consumption plan scales automatically and bills per execution, matching the stem's cost-effectiveness and variable-workload constraints. It triggers PDF processing jobs that call Azure OpenAI, avoiding idle capacity charges that always-on compute would incur.

Why this answer

Azure Functions with a consumption plan provides a serverless, event-driven architecture that scales automatically to handle variable workloads, ensuring cost-effectiveness by charging only for compute time used. This architecture is ideal for processing large volumes of PDFs, as each document can trigger a function execution that calls Azure OpenAI for summarization, without the need for always-on infrastructure.

Exam trap

Microsoft often tests the misconception that GPU or specialized compute is required for AI workloads, but in this scenario, the heavy lifting is done by Azure OpenAI's API, so the focus should be on cost-effective, scalable compute for orchestration, not local GPU processing.

How to eliminate wrong answers

Option A is wrong because Azure Kubernetes Service (AKS) with a persistent node pool of GPU nodes incurs continuous costs even during idle periods, making it less cost-effective for variable workloads, and the GPU nodes are unnecessary since Azure OpenAI is called via API, not run locally. Option C is wrong because deploying a GPU-enabled virtual machine and running summarization jobs sequentially introduces a single point of failure, lacks auto-scaling, and wastes resources on GPU hardware that is not required for API calls. Option D is wrong because Azure Logic Apps is designed for workflow orchestration and integration, not for high-throughput, cost-effective batch processing of large document volumes, and it would incur higher costs per execution compared to Azure Functions.

Full explanation →

459

MCQeasy

You are planning a solution that uses Azure AI Language to analyze customer feedback from social media posts. The solution must: - Detect sentiment (positive, negative, neutral) for each post. - Extract key phrases. - Support English and Spanish languages. - Run asynchronously for a batch of 10,000 posts. - Use the least expensive option that meets requirements. What should you do?

A.Use the Azure AI Language service with the built-in sentiment analysis and key phrase extraction capabilities. Process posts in batches using the async API.

B.Build a custom text classification model in Azure AI Language to detect sentiment and extract key phrases.

C.Use the Azure AI Language service with the single-document API for each post.

D.Use Azure AI Translator to translate all posts to English, then use Azure AI Language for analysis.

AnswerA

The built-in sentiment analysis and key phrase extraction capabilities cover both required tasks and support English and Spanish. The async API handles the 10,000-post batch, and using built-in features avoids the higher cost of custom models.

Why this answer

Azure AI Language's built-in sentiment analysis and key phrase extraction natively support both English and Spanish, and the async batch API is designed for high-volume processing (e.g., 10,000 posts) at a lower cost than per-document calls. This approach meets all requirements without custom models or translation overhead.

Exam trap

The trap here is that candidates often assume custom models are required for multilingual support or that translation is necessary, when in fact Azure AI Language's built-in capabilities already cover English and Spanish natively.

How to eliminate wrong answers

Option B is wrong because building a custom text classification model is unnecessary and more expensive; the built-in capabilities already handle sentiment and key phrase extraction for the required languages. Option C is wrong because using the single-document API for each of 10,000 posts would incur higher costs and slower performance compared to the async batch API, which is designed for bulk processing. Option D is wrong because translating all posts to English adds unnecessary cost and latency, and Azure AI Language already supports Spanish natively for both sentiment analysis and key phrase extraction.

Full explanation →

460

MCQmedium

You are building a solution to extract key information from invoices using Azure AI Document Intelligence. The invoices contain fields such as invoice number, date, total amount, and line items. However, the model is not correctly extracting the line items. Which prebuilt model should you use?

A.Prebuilt-receipt model

B.Prebuilt-idDocument model

C.Prebuilt-invoice model

D.Prebuilt-layout model

AnswerC

The prebuilt-invoice model is trained to extract invoice-specific fields including line items, invoice number, date and total amount. Custom or general models lack this schema, so the prebuilt invoice model directly addresses the failing line-item extraction described in the scenario.

Why this answer

The prebuilt-invoice model is specifically trained to extract key fields from invoices, including invoice number, date, total amount, and line items. Unlike other prebuilt models, it has dedicated field extraction for line item details such as description, quantity, unit price, and total, making it the correct choice for this scenario.

Exam trap

The trap here is that candidates may confuse the prebuilt-layout model's ability to extract table structure with the prebuilt-invoice model's trained field extraction for invoice-specific data, leading them to choose option D thinking layout analysis is sufficient for line item extraction.

How to eliminate wrong answers

Option A is wrong because the prebuilt-receipt model is optimized for receipt documents, which typically lack structured line items with descriptions and unit prices found in invoices. Option B is wrong because the prebuilt-idDocument model is designed for identity documents like passports and driver's licenses, not financial documents with line items. Option D is wrong because the prebuilt-layout model extracts text and table structure but does not have trained field extraction for invoice-specific fields like line items, invoice number, or total amount.

Full explanation →

461

MCQmedium

A company is building a solution to analyze customer reviews images using Azure AI Vision. They need to extract text from images that may contain both printed and handwritten text. Which feature should they use?

A.Custom Vision

B.OCR API (optical character recognition)

C.Read API

D.Azure AI Document Intelligence

AnswerC

The Read API uses the OCR engine that handles both printed and handwritten text within the same image, satisfying the mixed-content constraint. Legacy OCR and handwriting-only endpoints cannot process both simultaneously, so Read is the only feature meeting this requirement.

Why this answer

The Read API is the correct choice because it is specifically designed to extract text from images containing both printed and handwritten text, using advanced OCR capabilities that support mixed content. Unlike the OCR API, which is optimized for printed text only, the Read API leverages deep learning models to handle varied handwriting styles and complex layouts, making it ideal for analyzing customer review images.

Exam trap

The trap here is that candidates confuse the OCR API with the Read API, assuming both handle handwritten text equally, but the OCR API is limited to printed text while the Read API is the only one that natively supports mixed printed and handwritten content.

How to eliminate wrong answers

Option A is wrong because Custom Vision is a service for training custom image classification and object detection models, not for text extraction. Option B is wrong because the OCR API (optical character recognition) is optimized for printed text and does not reliably extract handwritten text, which is a key requirement. Option D is wrong because Azure AI Document Intelligence (formerly Form Recognizer) is designed for structured document processing (e.g., forms, invoices) and is not the primary service for general text extraction from images with mixed printed and handwritten content.

Full explanation →

462

MCQhard

You have a real-time video processing pipeline using Azure AI Video Indexer. You need to detect when a specific person appears in archived video footage. Which approach minimizes latency and cost?

A.Use Video Indexer's face detection and indexing, then search

B.Extract keyframes and use Custom Vision to detect the person

C.Run face detection on every frame using Azure AI Face and store results

D.Use Azure AI Vision to detect faces in video frames and compare against a database

AnswerA

Video Indexer indexes faces once during ingestion, storing face identifiers alongside timestamps, so later searches match against that index rather than re-processing footage. This avoids repeated full-video analysis, minimising both compute cost and query latency for archived footage.

Why this answer

Video Indexer's built-in face detection and indexing automatically identifies and tracks faces during the indexing process, storing the results in a searchable metadata index. To detect when a specific person appears, you can then search the indexed metadata for that person's face ID or name, which avoids re-processing the video and minimizes both latency and cost. This approach leverages the one-time indexing cost and optimized search capabilities rather than running additional AI services on every frame.

Exam trap

The trap here is that candidates often assume Custom Vision or Azure AI Face are needed for custom person detection, overlooking that Video Indexer already provides built-in face detection and search capabilities that are optimized for archived video analysis.

How to eliminate wrong answers

Option B is wrong because extracting keyframes and using Custom Vision requires training a custom model and processing only keyframes, which may miss the person if they appear between keyframes, and the custom training adds overhead and cost without leveraging Video Indexer's built-in face indexing. Option C is wrong because running face detection on every frame using Azure AI Face would incur high compute and API costs per frame, and storing all results creates unnecessary data volume, making it far more expensive and slower than using Video Indexer's pre-indexed search. Option D is wrong because using Azure AI Vision to detect faces in video frames and comparing against a database requires frame-by-frame processing and external database lookups, which introduces latency and cost that Video Indexer's integrated indexing and search avoids.

Full explanation →

463

Multi-Selectmedium

You are building an Azure AI Search enrichment pipeline that processes PDF documents from Azure Blob Storage. The documents contain both text and images. You need to extract text from the images and also detect the language of the extracted text to route documents to language-specific processing. Which two built-in skills should you include in the skillset? (Choose two.)

Select 2 answers

A.Text Translation skill

B.OCR skill

C.Language Detection skill

D.Entity Recognition skill

E.Key Phrase Extraction skill

AnswersB, C

The OCR skill is a built-in cognitive skill that extracts text from images. It is essential for this scenario because the PDFs contain images with embedded text. The OCR skill can process image files or images extracted from PDFs (when using the Document Extraction skill). It outputs text that can then be used by other skills, such as language detection. Without OCR, the text in images would not be available for indexing or further enrichment.

Why this answer

The OCR skill is required to extract text from images within the PDFs. The Language Detection skill is required to identify the language of the extracted text for routing. Together, they enable the pipeline to process image-based text and then determine its language.

The other skills do not provide these capabilities: key phrase extraction, entity recognition, and translation serve different purposes.

Exam trap

The trap here is assuming that language detection can work directly on images or that OCR also detects language, when in fact they are separate steps that must be chained.

Full explanation →

464

MCQeasy

You are using Microsoft Copilot Studio to create an agent that helps users reset their passwords. The agent should first verify the user's identity using multi-factor authentication (MFA) before proceeding. Which feature should you configure?

A.Add a variable to store the user's identity status

B.Configure Authentication settings to require Microsoft Entra ID authentication with MFA policy

C.Add a Power Automate flow that calls Microsoft Entra ID MFA

D.Use a 'Sign in' topic trigger from the customer channel

AnswerB

Configuring Authentication to require Microsoft Entra ID with an MFA policy forces identity verification before the agent proceeds, satisfying the pre-reset verification requirement. This leverages Entra ID's native conditional access rather than building custom verification logic.

Why this answer

Microsoft Copilot Studio allows you to configure Authentication settings directly on the agent, and by selecting 'Microsoft Entra ID' as the authentication provider, you can enforce an MFA policy that is already configured in your Entra ID tenant. This ensures that before the agent processes any password reset logic, the user must complete MFA, satisfying the identity verification requirement without custom code or flows.

Exam trap

The trap here is that candidates often think they need to build custom MFA logic (e.g., via Power Automate or variables) when the platform already provides a native, declarative way to enforce MFA through Authentication settings, leading them to over-engineer the solution.

How to eliminate wrong answers

Option A is wrong because simply adding a variable to store the user's identity status does not enforce MFA; it only tracks a state that must be set by some other mechanism, leaving the actual verification unaddressed. Option C is wrong because while a Power Automate flow could call Microsoft Entra ID MFA, this approach is unnecessarily complex and indirect—Copilot Studio's built-in Authentication settings natively support Entra ID with MFA policy enforcement, making a separate flow redundant and less reliable. Option D is wrong because a 'Sign in' topic trigger from the customer channel only initiates a sign-in prompt but does not guarantee that MFA is enforced; the actual MFA requirement must be configured in the Authentication settings of the agent, not just in a topic trigger.

Full explanation →

465

MCQmedium

You are building a mobile app that allows users to take a photo of a product and get detailed information. The app uses Azure AI Custom Vision to classify products. You need to ensure low latency for inference. What should you do?

A.Increase the number of training iterations

B.Use the Azure AI Vision API directly

C.Use Azure Front Door to cache results

D.Export the Custom Vision model as a TensorFlow model and run on-device

AnswerD

Exporting the Custom Vision model as TensorFlow and running inference on-device removes the network round trip to the Azure endpoint entirely, which is the dominant latency source for a mobile app. Local execution satisfies the low-latency constraint.

Why this answer

Exporting the Custom Vision model as a TensorFlow model and running it on-device eliminates network latency entirely. Inference happens locally on the mobile device, which provides the lowest possible latency for real-time classification, especially when network connectivity is poor or inconsistent.

Exam trap

The trap here is that candidates assume cloud-based solutions (like Azure Front Door or Vision API) are always faster, but Microsoft explicitly tests the understanding that on-device inference eliminates network latency and is the optimal choice for low-latency mobile scenarios.

How to eliminate wrong answers

Option A is wrong because increasing the number of training iterations improves model accuracy, not inference latency; latency is determined by model architecture and runtime environment, not training steps. Option B is wrong because using the Azure AI Vision API directly requires a network round-trip to Azure, which introduces higher latency compared to on-device inference, and it does not leverage the custom classification model you built. Option C is wrong because Azure Front Door caches HTTP responses at edge locations, but inference results are dynamic and user-specific (each photo is unique), so caching would rarely hit and cannot reduce the latency of the actual inference call.

Full explanation →

466

MCQeasy

You are planning to use Azure AI Vision to analyze images for a retail inventory management application. The solution must detect products on shelves and read expiration dates. Which two Azure AI Vision capabilities should you use?

A.Image Captioning

B.Object Detection

C.Face Detection

D.Optical Character Recognition (OCR)

AnswerB, D

Object Detection returns bounding boxes and labels for multiple distinct items within an image, which is what locating products on shelves requires. OCR reads text but cannot identify product instances, so object detection covers the shelf-detection half of the scenario.

Why this answer

Object Detection (B) is correct because it identifies and locates products on shelves by drawing bounding boxes around each detected item, which is essential for inventory tracking. Optical Character Recognition (OCR) (D) is correct because it extracts text from images, enabling the reading of expiration dates printed on product labels or packaging.

Exam trap

The trap here is that candidates may confuse Image Captioning with Object Detection, assuming a descriptive caption could identify products, or overlook OCR because they think expiration dates are purely numeric and can be handled by simpler methods, but Azure AI Vision's OCR is specifically designed for text extraction from images.

How to eliminate wrong answers

Option A is wrong because Image Captioning generates a natural language description of the entire image scene, not specific object locations or text extraction, so it cannot detect products on shelves or read expiration dates. Option C is wrong because Face Detection is designed to locate human faces in images, not products or text, and has no relevance to retail inventory management tasks.

Full explanation →

467

Multi-Selectmedium

A developer is deploying a custom text classification model in Azure AI Language. The model must be accessible via a REST API with low latency. Which TWO actions should the developer take?

Select 2 answers

A.Use the batch processing API

B.Export the model as a Docker container

C.Obtain the endpoint URL and primary key from Language Studio

D.Deploy the model to a real-time endpoint

E.Deploy to a test endpoint in the Azure portal

AnswersC, D

Retrieving the endpoint URL and primary key from Language Studio provides the authentication credentials and base address required to call the deployed model's REST API. Without these, requests cannot be authenticated, regardless of deployment type or latency characteristics.

Why this answer

Option D is correct because deploying the custom text classification model to a real-time endpoint in Azure AI Language exposes a synchronous REST API that returns predictions immediately, satisfying the low-latency requirement. Option C is correct because, to call that REST API, the developer must obtain the endpoint URL and the primary key (or a secondary key) from the project's deployment details in Language Studio, which are used in the Ocp-Apim-Subscription-Key header for authentication. Option A is incorrect because the batch processing API is designed for asynchronous, high-volume jobs and does not provide the low-latency synchronous responses required here.

Option B is incorrect because exporting the model as a Docker container is for on-premises or disconnected container deployment, not for exposing the model through the managed Azure AI Language REST endpoint. Option E is incorrect because a test endpoint in the Azure portal is intended for validation and does not provide the production-grade, low-latency real-time REST API needed for the application.

Exam trap

The trap here is that candidates often confuse batch processing with real-time inference, assuming that any API endpoint can provide low latency, or they mistakenly think exporting to a Docker container is the standard way to expose a model via REST in Azure.

Full explanation →

468

MCQmedium

You have a computer vision solution that analyzes security camera feeds to detect people and vehicles. The solution uses Azure AI Vision Spatial Analysis. You need to ensure compliance with privacy regulations by blurring detected faces. Which feature should you enable?

A.Use Azure AI Content Safety to filter faces

B.Post-process frames with Azure AI Face client SDK

C.Enable face detection and redact faces using Azure AI Video Indexer

D.Enable face blurring in the Spatial Analysis configuration

AnswerD

Face blurring in the Spatial Analysis configuration redacts detected faces in the video stream before storage or transmission, satisfying the stem's privacy compliance requirement. It operates within the spatial analysis pipeline itself, unlike separate Face service redaction applied after processing.

Why this answer

Azure AI Vision Spatial Analysis includes a built-in face blurring feature that can be enabled directly in the Spatial Analysis configuration. This allows you to automatically blur detected faces in the video feed at the edge or in the cloud, ensuring compliance with privacy regulations without requiring additional services or post-processing steps.

Exam trap

The trap here is that candidates may confuse Azure AI Video Indexer's face redaction capabilities with Spatial Analysis's real-time face blurring, or assume that a separate SDK or service is required for face blurring when it is actually a built-in configuration option in Spatial Analysis.

How to eliminate wrong answers

Option A is wrong because Azure AI Content Safety is designed to detect and filter harmful content (e.g., violence, hate speech) in text, images, and video, not to blur faces. Option B is wrong because post-processing frames with the Azure AI Face client SDK would require additional development effort and latency, and it is not a native feature of Spatial Analysis; the face blurring is already integrated into the Spatial Analysis pipeline. Option C is wrong because Azure AI Video Indexer is a separate service for extracting insights from video files (e.g., transcripts, faces, emotions) and does not provide real-time face blurring for live security camera feeds; it is not part of the Spatial Analysis solution.

Full explanation →

469

MCQmedium

You are building a chatbot using Azure AI Language and need to handle user intents that are not covered by the predefined intents. What should you implement?

A.Custom entities to capture unknown phrases

B.A fallback intent in the QnA Maker knowledge base

C.A 'None' intent in a conversational language understanding project

D.A prebuilt intent from the LUIS catalog

AnswerC

A 'None' intent in conversational language understanding captures utterances matching no predefined intent, satisfying the requirement to handle uncovered user intents. Microsoft Entra ID is unrelated here; the mechanism is intent classification fallback, where unmatched utterances route to 'None' so the chatbot can respond gracefully rather than misclassifying them.

Why this answer

In a Conversational Language Understanding (CLU) project, the 'None' intent is specifically designed to capture utterances that do not match any of the defined intents. This intent acts as a catch-all for unrecognized user inputs, ensuring the chatbot can gracefully handle out-of-scope or ambiguous queries without misclassifying them into a predefined intent.

Exam trap

The trap here is that candidates often confuse the 'None' intent with a fallback mechanism in QnA Maker or assume that custom entities can substitute for intent handling, leading them to pick options that address different aspects of NLP processing rather than the specific requirement for unrecognized intents.

How to eliminate wrong answers

Option A is wrong because custom entities are used to extract specific data points from utterances, not to handle unrecognized intents; entities do not define intent classification behavior. Option B is wrong because QnA Maker is a separate service for FAQ-style question answering, not for intent recognition; a fallback intent in QnA Maker would only apply to unanswered QnA pairs, not to intents in a CLU project. Option D is wrong because prebuilt intents from the LUIS catalog are domain-specific (e.g., 'BookFlight') and cannot cover all possible out-of-scope user inputs; they are designed for common scenarios, not as a generic fallback.

Full explanation →

470

MCQhard

Your Azure AI Search solution uses a custom skill to call an external API. The skill runs locally but fails when deployed to the search service. What is the most likely cause?

A.The skill's output field mappings are missing.

B.The skill's input field mappings are incorrect.

C.The indexer name is misspelled in the skillset.

D.The skill endpoint is not publicly accessible via HTTPS.

AnswerD

Custom skills execute server-side within Azure AI Search, so the external API must be reachable over public HTTPS; localhost or private endpoints work locally but fail once deployed. This satisfies the stem's deployment constraint, where the skill's endpoint becomes unreachable from the search service's network context.

Why this answer

When a custom skill runs locally but fails after deployment to Azure AI Search, the most common cause is that the skill's endpoint is not publicly accessible via HTTPS. Azure AI Search indexers execute skills in the cloud and must be able to reach the external API over the internet using a secure HTTPS connection; localhost or HTTP endpoints will fail.

Exam trap

The trap here is that candidates assume the skill logic is faulty (input/output mappings) rather than recognizing that the network connectivity and HTTPS requirement is the fundamental difference between local testing and cloud execution.

How to eliminate wrong answers

Option A is wrong because missing output field mappings would cause the skill to execute successfully but fail to write results to the index, not prevent the skill from running. Option B is wrong because incorrect input field mappings would cause the skill to receive wrong or missing data but would not prevent the skill from being invoked or the endpoint from being called. Option C is wrong because a misspelled indexer name would cause the indexer to fail to run, but the skillset itself would still be valid and the custom skill endpoint would be reachable; the error would occur at the indexer level, not the skill execution.

Full explanation →

471

Multi-Selecteasy

Which TWO of the following are best practices for securing Azure AI services?

Select 2 answers

A.Expose endpoints publicly to simplify client access.

B.Disable diagnostic logging to reduce data exposure.

C.Enable diagnostic settings to audit usage and detect anomalies.

D.Share API keys among multiple applications for simplicity.

E.Use managed identities to authenticate to Azure AI services.

AnswersC, E

Enabling diagnostic settings streams resource logs and metrics to Log Analytics, Storage or Event Hubs, giving the audit trail needed to spot anomalous calls against your Azure AI services. This directly satisfies the stem's security-monitoring requirement, since usage auditing and anomaly detection depend on that telemetry being captured and retained.

Why this answer

Option C is correct because enabling diagnostic settings on Azure AI services streams resource logs and metrics to destinations such as Log Analytics, Azure Storage, or Event Hubs, which supports auditing usage, monitoring for anomalies, and meeting compliance requirements. Option E is correct because managed identities let applications authenticate to Azure AI services via Microsoft Entra ID tokens, eliminating the need to store or rotate API keys in code or configuration. Option A is incorrect since publicly exposing endpoints increases the attack surface; access should be restricted with private endpoints, network ACLs, or Azure Private Link.

Option B is incorrect because disabling diagnostic logging removes the audit trail needed to detect misuse and investigate incidents. Option D is incorrect because sharing API keys across applications prevents per-app revocation and least-privilege scoping, so keys should be unique, stored in Key Vault, and rotated regularly.

Exam trap

The trap here is that candidates may think exposing endpoints publicly is acceptable for simplicity (Option A) or that sharing API keys is harmless (Option D), but Azure's security model emphasizes least privilege and credential isolation.

Full explanation →

472

MCQmedium

You are building a knowledge mining solution that indexes technical manuals in multiple languages. The solution must enable users to search in their native language and retrieve results in the same language. Which approach should you use?

A.Detect the language of the query using Azure AI Language and then use a generic analyzer

B.Translate all queries to English using Azure AI Translator before searching

C.Use a single non-language-specific analyzer like 'standard.lucene' for all documents

D.Use language-specific analyzers in the Azure AI Search index for each language

AnswerD

Language-specific analyzers apply per-field lexical processing, so each language's tokens are stemmed and indexed correctly. This satisfies the requirement that users search and retrieve results in their native language, since queries analysed with the matching language analyzer return same-language documents rather than mistranslated or poorly tokenised matches.

Why this answer

Azure AI Search supports language-specific analyzers (e.g., 'de.microsoft' for German, 'fr.microsoft' for French) that apply linguistic rules such as stemming, lemmatization, and stop-word removal tailored to each language. This ensures that queries and documents are processed in the same language, enabling users to search and retrieve results in their native language without translation loss.

Exam trap

The trap here is that candidates assume translation or generic analyzers are sufficient, overlooking that Azure AI Search's language-specific analyzers are designed to preserve linguistic integrity and meet the exact requirement of native-language search and retrieval without cross-language conversion.

How to eliminate wrong answers

Option A is wrong because detecting the query language and then using a generic analyzer (like 'standard.lucene') would ignore language-specific linguistic rules, leading to poor recall and precision for non-English text (e.g., German compound words or French accents). Option B is wrong because translating all queries to English introduces translation latency, potential semantic errors, and forces results to be returned in English, violating the requirement to retrieve results in the user's native language. Option C is wrong because a non-language-specific analyzer like 'standard.lucene' performs only basic tokenization and lowercasing, failing to handle language-specific morphology (e.g., stemming for Arabic or diacritics for Spanish), which degrades search quality.

Full explanation →

473

MCQeasy

You are deploying an agentic solution using Azure AI Agent Service. The agent needs to be invoked from a custom application using REST API calls. Which endpoint should you use to send a message to the agent?

A.POST /threads/{thread_id}/runs

B.POST /threads

C.POST /threads/{thread_id}/messages

D.GET /agents

AnswerC

Messages are added to an existing conversation thread, so POST /threads/{thread_id}/messages is the correct REST operation for sending user input to the agent. Thread creation happens first; the run is then started separately to process that message.

Why this answer

To send a message to an existing conversation thread in Azure AI Agent Service, you must use the POST /threads/{thread_id}/messages endpoint. This adds the user's message to the specified thread, which the agent can then process in a subsequent run. The REST API requires the thread to already exist, and messages are posted directly to that thread's resource.

Exam trap

The trap here is that candidates confuse the endpoint for sending a message with the endpoint for starting a run, mistakenly thinking that POST /threads/{thread_id}/runs both sends the message and invokes the agent, when in fact messages must be added separately before a run.

How to eliminate wrong answers

Option A is wrong because POST /threads/{thread_id}/runs is used to start a run (i.e., invoke the agent to process messages) on an existing thread, not to send a new message. Option B is wrong because POST /threads creates a new thread, but does not send a message; it only initializes the conversation container. Option D is wrong because GET /agents retrieves a list of available agents, not for sending messages.

Full explanation →

474

Multi-Selectmedium

Which THREE components are required to build a custom question answering solution using Azure AI Language?

Select 3 answers

A.A Language Understanding (LUIS) app

B.A project in Azure AI Language

C.An endpoint to query the knowledge base

D.An Azure AI Bot Service resource

E.A knowledge base with question and answer pairs

AnswersB, C, E

Custom question answering stores question-answer pairs, sources and trained model versions inside an Azure AI Language project. The project defines the task type and is the prerequisite container for importing sources and training the knowledge base.

Why this answer

Option B is correct because a custom question answering solution in Azure AI Language is created as a project (formerly a knowledge base) within the Azure AI Language resource, which holds the Q&A data and configuration. Option E is correct because the project must contain a knowledge base populated with question-and-answer pairs (or imported sources such as FAQs and documents) that the service indexes and searches. Option C is correct because after the knowledge base is built and deployed, a query endpoint is required so client applications can send questions and receive answers via the REST API or SDK.

Option A is not needed because LUIS is a separate language understanding service for intent and entity extraction, not for custom question answering. Option D is not required because Azure AI Bot Service is only an optional client that can consume the endpoint; the question answering solution itself does not depend on it.

Exam trap

AI-102 often tests the misconception that LUIS or Bot Service are required for question answering, confusing the separate Azure AI services and their roles.

Full explanation →

475

MCQmedium

You are a cloud solution architect at a legal firm. The firm needs to automate the summarization of legal documents. They have a large corpus of past case summaries and legal documents stored in Azure Blob Storage. They want to use Azure OpenAI to generate summaries for new documents. The solution must ensure that the generated summaries are accurate and do not contain hallucinated legal facts. The firm also requires that the solution be serverless and minimize operational overhead. You need to design the solution. Option A: Use Azure OpenAI with a system message that instructs the model to be accurate. Deploy the model as a web app on Azure App Service and call it from Azure Functions triggered by new blob uploads. Option B: Use Azure OpenAI with Retrieval-Augmented Generation (RAG) by indexing the past case summaries in Azure AI Search. Use Azure Functions to process new documents, retrieve relevant cases, and pass them as context to the model. Store summaries in Azure Cosmos DB. Option C: Fine-tune an Azure OpenAI model on the past case summaries and deploy it as a managed endpoint. Use Azure Logic Apps to trigger summarization when new blobs are added. Option D: Use Azure OpenAI with the chat API and provide the entire document in the prompt. Use Azure Container Instances to run a service that calls the API and writes summaries back to Blob Storage. Which option should you choose?

A.Option B

B.Option A

C.Option D

D.Option C

AnswerA

RAG grounds responses in retrieved documents, reducing hallucination.

Why this answer

It uses Retrieval-Augmented Generation (RAG) with Azure AI Search to ground the model's output in verified past case summaries, directly addressing the requirement to avoid hallucinated legal facts. The serverless architecture is achieved via Azure Functions triggered by blob uploads, minimizing operational overhead, while storing summaries in Azure Cosmos DB provides a scalable, low-latency output store.

Exam trap

The trap here is that candidates may assume fine-tuning (Option C) or a simple system message (Option A) is sufficient to ensure factual accuracy, but Azure OpenAI models require grounded context via RAG to reliably avoid hallucination in domain-specific tasks like legal summarization.

How to eliminate wrong answers

Option A is wrong because a system message alone cannot prevent hallucination; the model may still fabricate legal facts without grounded context. Option C is wrong because fine-tuning on past case summaries does not guarantee factual accuracy for new, unseen documents and introduces operational overhead with a managed endpoint, contradicting the serverless requirement. Option D is wrong because providing the entire document in the prompt without retrieval augmentation does not anchor the model to verified facts, and Azure Container Instances adds operational overhead compared to a serverless trigger.

Full explanation →

476

MCQmedium

You are deploying a Conversational Language Understanding (CLU) model to production. You need to monitor the model's performance and detect when retraining is needed due to concept drift. Which metric should you monitor?

A.Response time for each prediction

B.Number of endpoint calls

C.Number of utterances processed per day

D.Average confidence scores of predictions

AnswerD

Average confidence scores of predictions reveal declining certainty as utterances drift from training data, indicating concept drift. Monitoring this metric satisfies the requirement to detect when retraining is needed, since sustained low confidence signals the model no longer matches production input.

Why this answer

Average confidence scores of predictions is the correct metric because a sustained drop in confidence indicates that the model is encountering utterances that differ from its training distribution, which is a classic sign of concept drift. Monitoring confidence scores allows you to detect when the model's predictions become less certain, triggering the need for retraining with new data.

Exam trap

The trap here is that candidates confuse operational metrics (like response time or throughput) with model performance metrics, assuming any change in usage patterns indicates drift, when in fact only a drop in prediction confidence directly reflects model uncertainty.

How to eliminate wrong answers

Option A is wrong because response time measures latency, not prediction quality or drift; it can be affected by infrastructure issues but does not indicate whether the model's understanding has degraded. Option B is wrong because the number of endpoint calls reflects usage volume, not the accuracy or relevance of predictions; high traffic does not imply drift. Option C is wrong because the number of utterances processed per day is a throughput metric that shows how much data is being handled, but it does not reveal whether the model's performance on that data has declined.

Full explanation →

477

MCQmedium

A company is building a chatbot using Azure OpenAI Service to answer customer queries. The chatbot must not generate harmful or offensive content. Which Azure AI service should be integrated to filter inappropriate content?

A.Azure Bot Service

B.Azure Cognitive Search

C.Azure AI Content Safety

D.Azure Form Recognizer

AnswerC

Azure AI Content Safety provides dedicated moderation models that detect harmful, violent, hateful and sexual content in both prompts and completions. Integrating it filters inappropriate generated output, satisfying the requirement that the chatbot must not produce offensive responses.

Why this answer

Azure AI Content Safety is the correct service because it provides built-in content moderation APIs that detect and filter harmful or offensive text and images, including hate speech, violence, self-harm, and sexual content. Integrating this service with the Azure OpenAI chatbot ensures that user inputs and model outputs are screened in real time, preventing the generation of inappropriate responses.

Exam trap

The trap here is that candidates often confuse Azure Bot Service's ability to 'manage conversations' with built-in content filtering, but it actually lacks native moderation and requires explicit integration with a dedicated content safety service.

How to eliminate wrong answers

Option A is wrong because Azure Bot Service is a framework for building, deploying, and managing bots, but it does not include native content filtering capabilities; it would require integration with a separate content moderation service. Option B is wrong because Azure Cognitive Search is used for indexing and searching over structured and unstructured data, not for filtering harmful content in real-time chat interactions. Option D is wrong because Azure Form Recognizer (now Azure AI Document Intelligence) is designed to extract information from forms and documents, not to moderate or filter offensive language or imagery.

Full explanation →

478

Matchingmedium

Match each Azure AI service to its primary function.

Drag a concept onto its matching description — or click a concept then click the description.

Concepts

Matches

Build conversational AI bots

AI-powered cloud search

Extract information from documents

Analyze video and audio content

Monitor metrics and detect anomalies

Why these pairings

The correct matches pair each service with its primary function. Computer Vision analyzes images/videos; Language Service processes text; Speech Service handles audio; Cognitive Search provides AI search. Common confusions include swapping Speech with Computer Vision or Language with Cognitive Search.

Full explanation →

479

MCQmedium

You are an Azure AI engineer at Fabrikam Inc. The company has developed a custom vision model using Azure Custom Vision to detect defects on a manufacturing assembly line. The model is deployed as a Docker container to an on-premises edge device using Azure IoT Edge. Recently, the model's inference accuracy has decreased. The operations team reports that the edge device is running low on memory and CPU. The model was trained with images from a specific camera angle, but the camera angle has been changed slightly due to maintenance. You need to improve the model's accuracy. What should you do?

A.Upgrade the edge device to have more memory and CPU.

B.Reduce the image resolution to lower memory usage.

C.Retrain the model with new images captured from the current camera angle.

D.Convert the model to use grayscale images.

AnswerC

The camera angle shift changed the input distribution, so the model now infers on images unlike its training data. Retraining with images captured from the current angle realigns the model with production input, addressing the accuracy drop at its source.

Why this answer

The decrease in accuracy is most likely due to the change in camera angle, which introduces a domain shift between the training images and the new inference images. Retraining the model with images captured from the current camera angle will realign the training data distribution with the production environment, directly addressing the root cause of the accuracy drop. This is a standard practice in Custom Vision when deployment conditions change.

Exam trap

The trap here is that candidates focus on the resource constraints (low memory/CPU) as the primary cause of accuracy loss, but the question explicitly states the camera angle changed, making retraining the only option that addresses the domain shift.

How to eliminate wrong answers

Option A is wrong because upgrading hardware (more memory/CPU) addresses resource constraints but does not fix the accuracy degradation caused by the camera angle change; the model's inference logic remains unchanged. Option B is wrong because reducing image resolution may lower memory usage but will likely further degrade accuracy by removing fine-grained defect details, and it does not correct the domain shift from the new camera angle. Option D is wrong because converting to grayscale discards color information that may be critical for defect detection (e.g., color-based anomalies), and it does not address the camera angle change.

Full explanation →

480

MCQhard

A healthcare startup is developing a chatbot that uses Azure OpenAI to answer patient questions. They need to ensure that the chatbot only uses information from their verified medical database and does not generate unsupported medical advice. What is the best approach?

A.Fine-tune a model on the medical database and deploy it.

B.Embed the entire medical database in the system message.

C.Rely on Azure OpenAI's content filtering to block unsupported advice.

D.Use Azure AI Search with vector search to retrieve relevant documents and pass them as context.

AnswerD

Retrieval-augmented generation grounds responses in your own data: Azure AI Search with vector search returns the most semantically relevant verified medical documents, which are injected into the prompt as context. This constrains the model to the medical database, preventing unsupported advice.

Why this answer

It uses Azure AI Search with vector search to retrieve only relevant, verified documents from the medical database and passes them as context to the Azure OpenAI model. This grounds the model's responses in authoritative data, preventing it from generating unsupported medical advice. The retrieval-augmented generation (RAG) pattern ensures the chatbot answers are based on the provided context rather than the model's internal knowledge.

Exam trap

Microsoft often tests the misconception that fine-tuning or content filtering alone can control factual accuracy, when in reality retrieval-augmented generation (RAG) with Azure AI Search is the correct pattern for grounding responses in specific, verified data.

How to eliminate wrong answers

Option A is wrong because fine-tuning a model on a medical database does not guarantee it will avoid generating unsupported advice; the model can still hallucinate or produce information not present in the training data, and fine-tuning does not enforce retrieval of specific verified documents at inference time. Option B is wrong because embedding the entire medical database in the system message would exceed the token limit (typically 4,096 or 8,192 tokens for most models), making it impractical and inefficient, and it would not allow dynamic retrieval of the most relevant information. Option C is wrong because Azure OpenAI's content filtering is designed to block harmful or offensive content, not to verify the factual accuracy or medical validity of the model's responses; it cannot prevent the generation of unsupported medical advice that appears plausible.

Full explanation →

481

MCQmedium

You are building a generative AI assistant with Azure OpenAI Service that must answer questions using a large corpus of internal product manuals. The manuals are updated weekly, and the assistant must reflect changes without retraining the model. You need to implement a retrieval-augmented generation (RAG) pattern. What should you use to index and retrieve relevant manual content?

A.Use Azure Cosmos DB with the built-in vector search and connect it directly to the model via a function calling tool.

B.Create an Azure AI Search index with vector embeddings of the manual chunks and use it as a data source in the Azure OpenAI 'on your data' configuration.

C.Store the manuals as plain text in Azure Blob Storage and pass the entire corpus in each prompt.

D.Fine-tune the base model on the manuals each week using Azure OpenAI fine-tuning.

AnswerB

This correctly implements RAG by using vector search to retrieve semantically relevant manual chunks, which are then passed to the model. Azure AI Search supports vector indexes and integrates with Azure OpenAI 'on your data', enabling weekly updates without retraining. It is the standard, supported approach for grounding generative responses in dynamic internal content.

Why this answer

The correct approach is to use Azure AI Search with vector embeddings as a data source for Azure OpenAI 'on your data'. This enables retrieval-augmented generation, where relevant manual chunks are retrieved and supplied to the model, ensuring responses reflect weekly updates without retraining. It is the supported, scalable method for grounding generative AI in dynamic internal documents.

Exam trap

The trap here is assuming that fine-tuning or passing all documents in the prompt can substitute for a proper retrieval index.

Full explanation →

482

MCQeasy

A retail company is creating an Azure AI Foundry agent that helps customers find products. The agent must be able to call a product search API and a store inventory API. The team wants to define these capabilities so the agent can invoke them when needed. What should the team do?

A.Embed the API endpoints in the agent's system message and instruct the model to call them by using HTTP requests.

B.Use the agent's built-in code interpreter to write Python code that calls the APIs.

C.Create an Azure Function for each API and configure the agent to use them as skills.

D.Add the two APIs as tools in the agent definition, providing their OpenAPI schemas.

AnswerD

Azure AI Foundry agents use tools to interact with external services. By adding the APIs as tools with their OpenAPI schemas, the agent can understand the available operations, parameters, and responses, and invoke them when appropriate. This is the standard way to extend an agent's capabilities with custom APIs.

Why this answer

The team needs to enable the agent to call two external APIs. In Azure AI Foundry, tools are the mechanism for integrating external services. By adding each API as a tool with its OpenAPI schema, the agent gains the ability to invoke them with correct parameters and handle responses.

This is the native, supported approach and requires minimal custom code.

Exam trap

The trap here is assuming the agent can call APIs by simply mentioning them in the prompt, rather than defining them as tools with proper schemas.

Full explanation →

483

Multi-Selecthard

You are deploying a retrieval-augmented generation chat solution on Azure OpenAI. During testing, users report that the model confidently answers questions using facts that are not in the indexed documents, and that retrieved chunks sometimes come from the wrong department's SharePoint site. You need to reduce these ungrounded answers without retraining the model. (Choose two.)

Select 2 answers

A.Set the temperature parameter to 2 to make the model more factual.

B.Increase the max_tokens parameter so the model has more room to explain its reasoning before answering.

C.Set the message role "system" content to instruct the model to answer only from the provided context and to state when the answer is not found.

D.Enable content filtering at the highest severity threshold to block ungrounded statements.

E.Apply security trimming and metadata filters so retrieval only returns chunks the user is allowed to see and that match the requested department.

AnswersC, E

A system message sets behavioral boundaries for the whole chat session, so telling the model to restrict itself to supplied context and to admit when it cannot find an answer directly reduces fabricated responses. This is the standard, non-retraining grounding control used with Azure OpenAI chat completions, and it works alongside retrieval quality fixes rather than replacing them.

Why this answer

Ungrounded answers in a RAG pattern usually come from two sources: the model not being constrained to the retrieved context, and the retrieval step returning irrelevant or unauthorized chunks. A system message that enforces answer-only-from-context behavior, combined with metadata filtering and security trimming at the search layer, addresses both causes without retraining. Length and randomness parameters, plus harmful-content filters, do not influence factual grounding.

Exam trap

The trap here is assuming that lowering or raising sampling parameters such as temperature or max_tokens will fix hallucinations, when grounding depends on prompt constraints and retrieval quality instead.

Full explanation →

484

MCQmedium

Refer to the exhibit. You are creating a Custom Vision project using the Azure AI Custom Vision API. The training data is stored in Azure Blob Storage with a SAS URI. The project creation fails with an authorization error. What is the most likely reason?

A.The domain 'general' is invalid

B.The exportModelContainerUri is missing

C.The SAS token is expired or has insufficient permissions

D.The project type 'Classification' is not supported

AnswerC

The Custom Vision API reads training images from the blob container using the supplied SAS URI. If that token has expired or lacks read and list permissions on the container, the project creation request fails authorisation, which matches the reported error exactly.

Why this answer

The Custom Vision project creation fails because the SAS URI used to access training data in Azure Blob Storage has an expired token or lacks sufficient permissions (e.g., read/list). Custom Vision requires a valid SAS token with at least read and list permissions to import images from the container. An expired or under-permissioned SAS token results in an authorization error when the service attempts to access the blob storage.

Exam trap

The trap here is that candidates may confuse a SAS authorization error with a missing container URI or an invalid domain, when in fact the SAS token's expiry or insufficient permissions is the direct cause of the 403 error during blob access.

How to eliminate wrong answers

Option A is wrong because 'general' is a valid and commonly used domain for Custom Vision projects; it is not invalid. Option B is wrong because exportModelContainerUri is only required when exporting a trained model to a container, not for project creation or importing training data. Option D is wrong because 'Classification' is a fully supported project type in Custom Vision; the error is authorization-related, not about unsupported project types.

Full explanation →

485

MCQmedium

Your organization is using Azure AI Document Intelligence to process expense reports. The reports are submitted as images and need to be classified into categories (e.g., travel, office supplies) before extraction. Which feature of Document Intelligence should you use?

A.Custom classification model

B.OCR capability

C.Layout extraction

D.Prebuilt expense report model

AnswerA

A custom classification model trains on labelled samples to categorise documents into your own classes, such as travel or office supplies, before extraction. This satisfies the requirement to classify expense report images into categories prior to extracting fields.

Why this answer

Azure AI Document Intelligence's custom classification model is specifically designed to categorize documents (such as expense report images) into user-defined classes (e.g., travel, office supplies) before any extraction occurs. This model uses a trained classifier to assign a document type based on its visual and textual features, enabling downstream processing with the appropriate extraction model.

Exam trap

The trap here is that candidates often confuse the prebuilt expense report model (which extracts data) with the classification model (which categorizes documents), leading them to select Option D despite the question explicitly asking for classification before extraction.

How to eliminate wrong answers

Option B is wrong because OCR (Optical Character Recognition) capability only extracts text from images and does not perform document classification or categorization. Option C is wrong because layout extraction analyzes the structure (tables, paragraphs, headers) of a document but does not assign it to a predefined category. Option D is wrong because the prebuilt expense report model is designed to extract fields (e.g., vendor, total) from a known expense report format, not to classify arbitrary submitted images into categories like travel or office supplies.

Full explanation →

486

MCQhard

A company uses the Face API to detect and identify employees for building access. They need to ensure that the system complies with GDPR requirements for biometric data. Which action should they take?

A.Store faces in a secure database and delete after 30 days.

B.Anonymize the face data by blurring key features.

C.Obtain explicit consent from each employee before enrollment.

D.Use encryption for stored face templates.

AnswerC

Biometric templates derived from facial images are special-category personal data under GDPR, requiring a lawful basis beyond legitimate interest. Explicit, freely given consent from each employee before enrolment satisfies Article 9, and employees must be able to withdraw it without detriment.

Why this answer

Under GDPR, biometric data (such as facial recognition templates) is classified as special category data requiring explicit consent for processing. The Face API itself does not manage consent; the responsibility lies with the application layer. Option C is correct because obtaining explicit consent from each employee before enrollment is a fundamental GDPR requirement for lawful processing of biometric data.

Exam trap

The trap here is that candidates often focus on technical security measures (encryption, deletion, anonymization) as sufficient for GDPR compliance, overlooking the foundational legal requirement for explicit consent when processing special category biometric data.

How to eliminate wrong answers

Option A is wrong because merely storing faces in a secure database and deleting after 30 days does not address the GDPR requirement for a lawful basis (e.g., explicit consent) before processing biometric data; retention limits are a separate compliance aspect. Option B is wrong because anonymizing face data by blurring key features would render the Face API ineffective for identification, as the API requires clear facial features to generate a unique face template; this approach would break the system's core functionality. Option D is wrong because encryption of stored face templates protects data at rest but does not provide a lawful basis for processing; GDPR requires a valid legal ground (such as explicit consent) regardless of encryption.

Full explanation →

487

MCQhard

You are designing an Azure AI Search enrichment pipeline that extracts entities from text using the Entity Recognition skill. You need to ensure that the extracted entities are stored as a collection in the index so that users can filter and facet on them. Which index field type should you use?

A.Edm.ComplexType

B.Edm.String

C.Edm.Int32

D.Collection(Edm.String)

AnswerD

Collection(Edm.String) allows storing multiple string values in a single field. This is ideal for entities extracted by the Entity Recognition skill, which outputs a list of entities. With this type, you can apply filters and facets on each entity value. It preserves the collection nature of the output and enables advanced query scenarios.

Why this answer

The Entity Recognition skill outputs a list of strings representing entities. To store this list and enable filtering and faceting, the index field must be a collection of strings. Collection(Edm.String) is the correct type because it accommodates multiple values and supports the necessary query capabilities.

Other types either cannot hold multiple values or are not designed for text data.

Exam trap

The trap here is assuming that a complex type is needed to store structured entity data, when the skill actually outputs a simple string collection that can be directly mapped.

Full explanation →

488

MCQeasy

You are building a knowledge mining solution using Azure AI Search. You need to ensure that sensitive information such as credit card numbers is automatically removed from the indexed content. Which built-in skill should you add to your skillset?

A.Entity Recognition skill

B.Conditional skill

C.PII Detection skill

D.Text Translation skill

AnswerC

The PII Detection skill scans enriched content and identifies entities such as credit card numbers, then allows masking or redaction before the data is written to the index. This satisfies the requirement that sensitive information is automatically removed from indexed content during skillset execution.

Why this answer

The PII Detection skill is the correct built-in skill for automatically identifying and redacting sensitive information like credit card numbers from indexed content in Azure AI Search. It uses pre-trained models to detect patterns such as credit card numbers, social security numbers, and other personally identifiable information, and can either mask or remove them from the text before it is stored in the search index.

Exam trap

The trap here is that candidates may confuse the Entity Recognition skill (which can identify entities but not redact them) with the PII Detection skill, assuming that entity extraction inherently includes removal, when in fact redaction requires a separate skill designed for that purpose.

How to eliminate wrong answers

Option A is wrong because the Entity Recognition skill extracts named entities like people, organizations, and locations, but it does not have built-in redaction capabilities for sensitive data patterns like credit card numbers. Option B is wrong because the Conditional skill is used to apply conditional logic (if-then-else) to skill outputs, not to detect or remove sensitive information. Option D is wrong because the Text Translation skill translates text between languages and has no functionality for identifying or redacting sensitive data such as credit card numbers.

Full explanation →

489

MCQhard

You are designing a knowledge mining solution that must handle sensitive customer data. The solution must ensure that personally identifiable information (PII) is not returned in search results. What should you do?

A.Use Azure AI Search with encryption at rest

B.Implement role-based access control on the search index

C.Use a custom skill in the skillset to detect and redact PII before indexing

D.Configure field mappings to exclude PII fields

AnswerC

A custom skill runs PII detection and redaction during enrichment, so sensitive values are removed before documents reach the index. Because the index never stores the PII, search results cannot return it, satisfying the requirement.

Why this answer

A custom skill in the Azure AI Search skillset lets you invoke a custom function (e.g., Azure Function calling Azure AI Language's PII detection or a regex-based redactor) during the enrichment pipeline, before documents are indexed. This ensures PII is detected and redacted/removed at ingestion time, so it never appears in the search index or query results.

Exam trap

AI-102 often tests the confusion between access control (RBAC) and data minimization (redaction) — candidates pick RBAC thinking 'restricting access' satisfies 'not returned in results,' but the requirement is about the data itself, not who can see it.

How to eliminate wrong answers

Option A is wrong because encryption at rest protects data from unauthorized access to storage but does not prevent PII from being returned in search results — the data is still indexed and queryable. Option B is wrong because RBAC controls who can query the index, but authorized users would still see PII; the requirement is to prevent PII from being returned at all, not just to restrict access. Option D is wrong because field mappings only map source fields to index fields — they do not detect or redact PII within content; excluding a field by name assumes PII is isolated in known fields, which is not reliable for free-text documents.

Full explanation →

490

MCQeasy

You are planning to use Azure AI Document Intelligence to process a large volume of mixed document types (invoices, receipts, and purchase orders). The solution must automatically classify each document type and extract relevant fields. What should you configure?

A.Use the Form Recognizer service with neural models

B.Create a custom classification model to identify document types, then use extraction models

C.Use prebuilt models for each document type and route based on filename

D.Use the Read model to extract all text and then use regular expressions to classify

AnswerB

A custom classification model is trained on labelled samples of invoices, receipts and purchase orders, so it identifies each document's type before routing. Extraction models then run per type, satisfying the requirement to classify automatically and pull the relevant fields from mixed documents.

Why this answer

Azure AI Document Intelligence (formerly Form Recognizer) requires a two-step process for mixed document types: first, a custom classification model identifies each document type (invoice, receipt, purchase order), then separate extraction models (custom or prebuilt) extract the relevant fields from each classified type. This approach ensures accurate routing and field extraction without relying on filenames or brittle regex patterns.

Exam trap

The trap here is that candidates assume a single model (like neural or prebuilt) can both classify and extract, but Azure AI Document Intelligence requires a separate classification step before extraction for mixed document types.

How to eliminate wrong answers

Option A is wrong because neural models are a type of extraction model (for improved accuracy on complex documents) but do not provide document-type classification; they cannot automatically distinguish between invoices, receipts, and purchase orders without a separate classifier. Option C is wrong because routing based on filename is unreliable and not a supported feature of Document Intelligence; filenames can be inconsistent or missing, and the service requires explicit classification logic. Option D is wrong because the Read model only extracts raw text and layout, not structured fields, and using regular expressions to classify document types is error-prone and not scalable for mixed document types with varying formats.

Full explanation →

491

Multi-Selecthard

A company uses Azure AI Language to analyze customer reviews. They need to detect sentiment, extract key phrases, and identify named entities. Which THREE capabilities should they combine?

Select 3 answers

A.Named entity recognition (NER)

B.Abstractive summarization

C.Language detection

D.Key phrase extraction

E.Sentiment analysis

AnswersA, D, E

Named entity recognition identifies and categorises entities such as people, places, organisations and dates within review text, directly satisfying the requirement to identify named entities. It is one of the three Azure AI Language capabilities the solution must combine alongside sentiment analysis and key phrase extraction.

Why this answer

Sentiment analysis (E) is correct because it returns sentiment labels and confidence scores (positive, negative, neutral, mixed) for the review text, directly satisfying the requirement to detect sentiment. Key phrase extraction (D) is correct because it identifies the main talking points in unstructured text, which fulfills the requirement to extract key phrases. Named entity recognition (A) is correct because it detects and categorizes entities such as people, places, organizations, and dates, satisfying the requirement to identify named entities.

Abstractive summarization (B) is not needed because it generates new condensed text rather than performing the three requested analyses, and language detection (C) is not required because the scenario does not ask to identify the review language.

Exam trap

Microsoft often tests the distinction between core NLP capabilities (sentiment, NER, key phrase extraction) and advanced features like summarization or language detection, so candidates may mistakenly include abstractive summarization because it sounds like it could help analyze reviews, but it is not one of the three required capabilities.

Full explanation →

492

MCQmedium

You are building an application that must detect and redact personally identifiable information (PII) from free-text support tickets before storing them. The tickets are written in English, and you need to identify entities such as names, phone numbers, and email addresses, and replace them with asterisks. You plan to use the Azure AI Language service. Which API endpoint should you call?

A.POST /language/:analyze-text with kind set to "KeyPhraseExtraction" in the request body.

B.POST /language/:analyze-text with kind set to "PiiEntityRecognition" in the request body.

C.POST /language/:analyze-text with kind set to "EntityRecognition" in the request body.

D.POST /language/:analyze-conversations with kind set to "ConversationPII" in the request body.

AnswerB

The PiiEntityRecognition kind is specifically designed to detect and optionally redact personal data such as names, phone numbers, email addresses, and other PII categories. By default, the response includes redactedText where detected entities are replaced with asterisks, satisfying the requirement to redact before storage without additional processing.

Why this answer

The PiiEntityRecognition kind in the analyze-text endpoint is purpose-built for identifying and redacting personal data. It returns both the detected entities and a redactedText field where entities are masked with asterisks. The other kinds either detect different entity types or serve different NLP tasks, and none provide the required redaction behavior.

Exam trap

The trap here is confusing general entity recognition with PII detection, assuming that EntityRecognition also redacts personal data.

Full explanation →

493

Multi-Selectmedium

Which TWO Azure AI Search features should you enable to improve the relevance of search results for a knowledge mining solution that supports natural language queries?

Select 2 answers

A.Synonyms

B.Semantic ranking

C.Search mode 'all'

D.Scoring profiles

E.Filters

AnswersB, D

Semantic ranking applies Microsoft's language models to re-rank the initial result set, so natural language queries return contextually relevant matches rather than keyword-only hits. It satisfies the requirement to improve relevance for conversational queries in the knowledge mining solution.

Why this answer

Semantic ranking (B) is correct because it uses Microsoft's language understanding models to re-rank the top results from the initial BM25 retrieval, promoting results that are semantically relevant to natural language queries rather than just keyword matches. Scoring profiles (D) are correct because they let you boost or demote documents based on weighted fields, functions (e.g., magnitude, freshness, distance), and parameters, directly tuning relevance for the knowledge mining scenario. Synonyms (A) only expand query terms with equivalent expressions and do not provide semantic re-ranking or weighted relevance boosting.

Search mode 'all' (C) merely requires all query terms to match, which is a stricter boolean behavior that can reduce recall rather than improve relevance. Filters (E) restrict the result set by criteria such as OData expressions but do not affect ranking or relevance scoring.

Exam trap

AI-102 often tests the distinction between recall-improving features (synonyms, search mode) and relevance-ranking features (semantic ranking, scoring profiles), where candidates incorrectly select synonyms or filters for relevance improvement.

Full explanation →

494

Multi-Selectmedium

You are designing an Azure AI Vision solution that must detect and extract text from identity documents such as passports and driver's licenses. The solution must also identify the document type and extract key fields like name and date of birth. You need to choose the appropriate Azure AI service and features. (Choose two.)

Select 2 answers

A.Azure AI Vision Read API

B.Azure AI Document Intelligence with the prebuilt-idDocument model

C.Azure AI Vision Image Analysis with the read feature

D.Azure AI Document Intelligence with the prebuilt-receipt model

E.Azure AI Document Intelligence with a custom extraction model

AnswersB, E

The prebuilt-idDocument model in Azure AI Document Intelligence is specifically trained to analyze identity documents such as passports and driver's licenses. It extracts key fields like name, date of birth, document number, and expiration date, and it identifies the document type. This directly meets the requirement for detecting and extracting text and fields from identity documents.

Why this answer

The prebuilt-idDocument model in Azure AI Document Intelligence is specifically designed to analyze identity documents and extract fields like name and date of birth. A custom extraction model in the same service can also be trained to extract custom fields and classify document types when prebuilt models are insufficient. The Read API and Image Analysis read feature only extract text without structured field extraction, and the prebuilt-receipt model is for receipts, not identity documents.

Exam trap

The trap here is assuming that any OCR service can extract structured fields from identity documents, when only specialized Document Intelligence models provide that capability.

Full explanation →

495

MCQeasy

Your company wants to build a custom Copilot for customer support using Microsoft Copilot Studio. The support team needs to query a backend CRM system securely using the Copilot. Which authentication method should you configure for the custom connector to the CRM?

A.Microsoft Entra ID OAuth 2.0

B.API key authentication

C.Basic authentication

D.Client certificate authentication

AnswerA

Microsoft Entra ID OAuth 2.0 lets Copilot Studio authenticate to the CRM through delegated, token-based authorisation, so the support team queries CRM data securely without embedding credentials. This satisfies the stem's secure backend CRM authentication requirement.

Why this answer

Microsoft Entra ID OAuth 2.0 is the correct authentication method because it provides secure, token-based delegated access to the backend CRM system without exposing long-lived credentials. Copilot Studio connectors that require user context and secure resource access should use OAuth 2.0 with Microsoft Entra ID, which supports modern authentication flows like authorization code grant and can enforce conditional access policies.

Exam trap

The trap here is that candidates often choose API key authentication because it seems simpler, but Microsoft Copilot Studio connectors for secure backend systems require OAuth 2.0 to support user delegation and token lifecycle management, not static keys.

How to eliminate wrong answers

Option B (API key authentication) is wrong because API keys are static, shared secrets that lack user context, token expiration, and granular scoping, making them unsuitable for secure, delegated access to a CRM in a Copilot scenario. Option C (Basic authentication) is wrong because it transmits credentials in plaintext (Base64-encoded) over HTTP, which is insecure and violates modern security best practices; it also does not support token refresh or scoped permissions. Option D (Client certificate authentication) is wrong because while it provides strong mutual TLS authentication, it is typically used for server-to-server or machine-to-machine scenarios, not for user-delegated access from a Copilot that needs to act on behalf of a support agent.

Full explanation →

496

MCQmedium

You are building an Azure OpenAI solution that must generate product descriptions from a small set of 40 curated examples that demonstrate your brand voice. You want the model to imitate the style without changing the model weights or incurring the cost of a fine-tuning job. What should you do?

A.Set the temperature parameter to 2 and the top_p parameter to 0.1 on every request.

B.Upload the examples to an Azure AI Search index and enable semantic ranker on the index.

C.Include several of the curated examples directly in the system and user messages of each chat completion request.

D.Create a fine-tuning job in Azure OpenAI Studio using the 40 examples and deploy the resulting custom model.

AnswerC

Placing curated examples in the system and user messages is few-shot prompting, which conditions the model on your brand voice at inference time without modifying weights. It is the appropriate approach when you have a small set of examples and want to avoid the cost and latency of a fine-tuning job. The examples guide tone and structure for each response.

Why this answer

Few-shot prompting embeds a handful of curated examples in the prompt so the model mimics their tone, structure, and vocabulary on each call. Because the requirement is a small example set and no weight modification or fine-tuning cost, in-context examples are the correct mechanism. Sampling parameters and retrieval indexes do not substitute for style conditioning.

Exam trap

The trap here is assuming that any style customization requires fine-tuning, when a small example set is better handled with in-context few-shot prompting.

Full explanation →

497

Multi-Selecthard

You are designing a solution that uses Azure AI Document Intelligence to extract data from invoices. The solution must handle high throughput and process documents in batch. Which TWO configuration options should you use?

Select 2 answers

A.Use the synchronous API for each invoice

B.Train a custom model for invoice extraction

C.Provision a standard (S0) tier Document Intelligence resource

D.Implement batch processing using the async document analysis API

E.Store output directly in Azure Blob Storage without API calls

AnswersC, D

Provisioning a standard (S0) tier resource satisfies the high-throughput requirement, since the free (F0) tier enforces strict request and page limits unsuitable for batch workloads. S0 removes those throttling caps, letting the batch pipeline submit many concurrent invoice analyses without hitting quota ceilings.

Why this answer

Option C is correct because the standard (S0) tier is the production-grade Document Intelligence pricing tier that supports high request volumes and higher transactions-per-second throughput, whereas the free (F0) tier is limited to low-volume testing and would throttle a high-throughput batch workload. Option D is correct because the asynchronous document analysis API (e.g., POST to /documentModels/{modelId}:analyze followed by polling the Operation-Location header) is specifically designed for large-scale, batch processing of many documents without blocking on each request, which is required for high throughput. Option A is wrong because the synchronous API processes one document per blocking call and does not scale well for batch, high-throughput scenarios.

Option B is wrong because training a custom model addresses extraction accuracy for specialized document layouts, not throughput or batch processing capability. Option E is wrong because storing output in Blob Storage without API calls bypasses Document Intelligence entirely, so no extraction would occur.

Exam trap

The trap here is that candidates often confuse the synchronous API (which is simpler but not scalable) with the async API (which is designed for batch and high throughput), and they may also overlook that the free tier (F0) cannot handle production-level batch loads.

Full explanation →

498

MCQmedium

You are implementing a generative AI chat solution on Azure OpenAI Service. The solution must ground its answers in a large internal knowledge base and return inline citations that link back to the exact source chunks. You want to minimize custom code and use a managed retrieval pipeline. Which Azure OpenAI feature should you configure?

A.Use the On Your Data feature with an Azure AI Search index and set the citation output to include document references.

B.Deploy a fine-tuned model trained on the internal knowledge base and use it for chat completions.

C.Set the temperature parameter to a low value and enable content filtering on the deployment.

D.Create a system message that instructs the model to answer only from the knowledge base and to include source names.

AnswerA

On Your Data with Azure AI Search is the managed retrieval-augmented generation feature of Azure OpenAI. It chunks, indexes, and retrieves content from your data source and returns citations that reference the source documents, with minimal custom code. Setting citations on ensures answers link back to the retrieved chunks.

Why this answer

Grounding answers in private data with verifiable citations requires a retrieval pipeline. Azure OpenAI On Your Data with an Azure AI Search index provides managed ingestion, chunking, retrieval, and citation output, so answers link to the actual source chunks. Prompt instructions or model tuning alone cannot retrieve documents or guarantee accurate references.

Exam trap

The trap here is assuming that a prompt or fine-tuned model can reliably cite internal documents without a retrieval index.

Full explanation →

499

MCQmedium

A company is developing an agent that uses Azure AI Vision to analyze images uploaded by users. The agent must identify objects and read text in images. The team uses the Azure AI Vision API. During testing, the agent fails to read text from images with low contrast. What should the team do to improve optical character recognition (OCR) accuracy for such images?

A.Use a different OCR API from Azure Cognitive Services.

B.Pre-process the image to adjust contrast and brightness before calling the OCR API.

C.Train a custom OCR model using Azure Custom Vision.

D.Increase the confidence threshold for text detection.

AnswerB

Low-contrast images degrade the pixel gradients OCR relies on to segment characters. Adjusting contrast and brightness before the API call restores that separation, improving recognition accuracy without retraining or changing the Azure AI Vision service.

Why this answer

Azure AI Vision's OCR API performs best on images with sufficient contrast and brightness. Pre-processing the image (e.g., using OpenCV or PIL to adjust contrast and brightness) enhances text visibility, directly improving OCR accuracy for low-contrast images without changing the API or training a custom model.

Exam trap

The trap here is that candidates may assume Azure Custom Vision can be repurposed for OCR or that adjusting confidence thresholds can fix recognition accuracy, when in fact pre-processing the image is the standard approach to improve OCR results for low-quality inputs.

How to eliminate wrong answers

Option A is wrong because Azure AI Vision's Read API is already the dedicated OCR service within Azure Cognitive Services; switching to a different OCR API (e.g., Form Recognizer) would not inherently solve low-contrast issues and may add unnecessary complexity. Option C is wrong because Azure Custom Vision is designed for image classification and object detection, not for reading text; it cannot be trained to perform OCR. Option D is wrong because increasing the confidence threshold for text detection would filter out more detections, potentially missing low-confidence but correct text reads, and does not improve the underlying OCR engine's ability to recognize text in low-contrast images.

Full explanation →

500

MCQhard

You are an AI engineer at Contoso. Contoso has a Microsoft Copilot for Microsoft 365 deployment. They want to build a custom copilot in Microsoft Copilot Studio that can answer questions about their internal IT support knowledge base stored in a SharePoint Online document library. The knowledge base includes hundreds of PDF and Word documents. Requirements: - The copilot must only answer from the approved knowledge base documents. - Responses must be grounded in the documents and include citations. - The solution must use generative answers with a prebuilt AI model (no custom model training). - Authentication must be via Microsoft Entra ID with single sign-on (SSO). - The copilot should be published to a Microsoft Teams channel. You need to recommend the minimal configuration steps. What should you do?

A.Create a custom Azure AI Language model using the documents. Then build a bot with Azure Bot Service and connect it to Copilot Studio.

B.Use Power Virtual Agents to create a bot. Add a custom entity to map document content. Use Power Automate to retrieve documents.

C.Use Azure AI Search to index the documents. Create a custom connector in Copilot Studio to query the search index. Configure authentication and publish to Teams.

D.In Copilot Studio, create a new copilot. Add the SharePoint document library as a knowledge source. Configure authentication with Microsoft Entra ID. Enable generative answers. Publish to the Teams channel.

AnswerD

Adding the SharePoint library as a knowledge source with generative answers grounds responses in approved documents and returns citations, while Microsoft Entra ID authentication delivers SSO. No custom model training is needed, and Teams publishing meets the channel requirement.

Why this answer

The minimal configuration is to create a copilot in Copilot Studio, add the SharePoint document library as a knowledge source, configure authentication with Microsoft Entra ID, enable generative answers, and publish to Teams. This uses built-in capabilities, requires no custom model training, and meets all requirements including grounding with citations and SSO.

Exam trap

AI-102 often tests the misconception that custom AI models or Azure AI Search are required for document Q&A, but Copilot Studio's native SharePoint knowledge source provides a no-code solution.

How to eliminate wrong answers

Option A is wrong because it involves creating a custom Azure AI Language model and Azure Bot Service, which is overkill and not minimal; it also may not support generative answers with citations as required. Option B is wrong because Power Virtual Agents (now part of Copilot Studio) with custom entities and Power Automate is not the recommended approach for document-based generative answers; it lacks the native knowledge source integration. Option C is wrong because using Azure AI Search with a custom connector adds unnecessary complexity and does not leverage the built-in SharePoint knowledge source in Copilot Studio, which automatically handles indexing and citations.

Full explanation →

501

MCQeasy

A company needs to implement a chatbot that answers customer queries using a knowledge base. Which Azure AI service should be used to build the knowledge base?

A.Azure AI Language (Question Answering)

B.Azure Bot Service

C.Azure AI Translator

D.Azure AI Speech

AnswerA

Azure AI Language's Question Answering feature builds a knowledge base from documents or FAQs and returns precise answers to natural-language queries, which the chatbot consumes. It directly satisfies the requirement to construct the knowledge base.

Why this answer

Azure AI Language's Question Answering feature is specifically designed to create a knowledge base from structured or unstructured content (e.g., FAQs, product manuals, support documents). It uses a custom question-answering model that can be trained and published as a REST API endpoint, which a chatbot can then query to retrieve precise answers. This makes it the correct service for building the knowledge base itself.

Exam trap

The trap here is that candidates often confuse the chatbot orchestration service (Azure Bot Service) with the knowledge base service itself, forgetting that the Bot Service is a host for the bot logic and channel integration, while the knowledge base must be built using a dedicated AI service like Question Answering.

How to eliminate wrong answers

Option B (Azure Bot Service) is wrong because it is a framework for building, deploying, and managing chatbots, not for creating or storing a knowledge base; it would consume a knowledge base built by another service. Option C (Azure AI Translator) is wrong because it provides real-time text translation between languages, not a question-answering knowledge base. Option D (Azure AI Speech) is wrong because it handles speech-to-text and text-to-speech capabilities, not the storage or retrieval of factual answers from a knowledge base.

Full explanation →

502

MCQeasy

A healthcare organization needs to redact personally identifiable information (PII) from patient records before using them for research. They have large volumes of unstructured text in multiple languages. Which Azure AI service should they use?

A.Azure AI Content Safety

B.Azure AI Language (PII detection feature)

C.Azure AI Document Intelligence (formerly Form Recognizer)

D.Azure AI Translator

AnswerB

Azure AI Language's PII detection handles unstructured text across multiple languages, satisfying both the multilingual and high-volume constraints in the stem. Its prebuilt NER models identify and redact entities such as names, addresses and medical identifiers, so patient records can be anonymised before research use without custom model training.

Why this answer

Azure AI Language's PII detection feature is specifically designed to identify and redact personally identifiable information from unstructured text. It supports multiple languages and can handle large volumes of text, making it the correct choice for redacting PII from patient records before research use.

Exam trap

The trap here is that candidates may confuse Azure AI Language's PII detection with Azure AI Document Intelligence's data extraction capabilities, but Document Intelligence focuses on structured field extraction from forms, not redaction of PII from unstructured text.

How to eliminate wrong answers

Option A is wrong because Azure AI Content Safety is used to detect harmful or offensive content (e.g., hate speech, violence), not to identify or redact PII. Option C is wrong because Azure AI Document Intelligence (formerly Form Recognizer) extracts structured data from documents (e.g., forms, invoices) but does not have native PII redaction capabilities for unstructured text. Option D is wrong because Azure AI Translator is a machine translation service that translates text between languages and does not include PII detection or redaction features.

Full explanation →

503

MCQmedium

A healthcare organization is building a clinical decision support system that must extract medical entities (e.g., symptoms, diagnoses, medications) from unstructured clinical notes. The solution must be able to detect relationships between entities, such as 'medication X treats symptom Y'. Which Azure AI service should be used?

A.Azure AI Translator

B.Azure AI Speech-to-Text

C.Azure AI Document Intelligence (formerly Form Recognizer)

D.Azure AI Language - Custom NER with entity linking

AnswerD

Azure AI Language's Custom NER with entity linking enables training a custom model to extract medical entities and link them to a knowledge base like UMLS, from which relationships can be inferred, making it the correct choice for this use case.

Why this answer

Azure AI Language's Custom Named Entity Recognition (NER) with entity linking allows training a model to extract domain-specific medical entities such as symptoms, diagnoses, and medications from unstructured clinical notes. Entity linking connects these entities to a knowledge base (e.g., UMLS), which contains structured relationships between medical concepts, enabling the inference of relationships like 'medication X treats symptom Y'. Thus, Custom NER with entity linking is the appropriate service for both entity extraction and relationship detection.

Exam trap

The trap here is that candidates may confuse Azure AI Document Intelligence's ability to extract text from forms with the need for custom entity extraction and relationship detection, overlooking that Document Intelligence lacks the natural language understanding capabilities required for unstructured clinical notes.

How to eliminate wrong answers

Option A is wrong because Azure AI Translator is designed for text translation between languages, not for extracting medical entities or detecting relationships between them from unstructured text. Option B is wrong because Azure AI Speech-to-Text converts spoken audio into text, but it does not perform entity extraction or relationship detection from the resulting text. Option C is wrong because Azure AI Document Intelligence (formerly Form Recognizer) is optimized for extracting structured data (e.g., key-value pairs, tables) from forms and documents, not for understanding complex medical relationships in unstructured clinical notes.

Full explanation →

504

MCQeasy

A team is building a chatbot using Azure Bot Service and Language Understanding (LUIS). The chatbot must handle multiple languages. What should you configure?

A.Configure a single LUIS app with multilingual utterances

B.Use Azure AI Search to index translated content

C.Use Azure Translator to translate utterances before calling LUIS

D.Create separate LUIS applications for each language

AnswerD

Separate LUIS applications per language satisfy the multilingual constraint because LUIS models are trained on a single language's utterances; a model cannot interpret intents across languages. Each app holds its own intents, entities and utterances, and the bot routes utterances to the matching app based on detected locale.

Why this answer

LUIS does not natively support multilingual models within a single app. Each LUIS application is designed for a single language, so to handle multiple languages, you must create separate LUIS apps—one per language—and route user utterances to the appropriate app based on the detected language. This ensures accurate intent and entity recognition tailored to each language's linguistic patterns.

Exam trap

The trap here is that candidates assume a single LUIS app can handle multiple languages by simply adding multilingual utterances, but LUIS explicitly requires separate apps per language because its models are language-specific and cannot generalize across languages.

How to eliminate wrong answers

Option A is wrong because a single LUIS app cannot be configured with multilingual utterances; LUIS apps are monolingual by design, and mixing languages in one app degrades prediction accuracy. Option B is wrong because Azure AI Search is a cognitive search service for indexing and querying content, not a tool for language detection or intent recognition in a chatbot; it does not replace the need for language-specific LUIS apps. Option C is wrong because while Azure Translator can translate utterances, translating before calling LUIS introduces latency, potential loss of nuance, and is not a recommended pattern—LUIS expects native-language utterances for optimal performance, and translation is better handled at the application layer if needed.

Full explanation →

505

MCQeasy

You are planning to deploy an Azure AI solution that uses Azure Cognitive Services. You need to ensure that the solution can be deployed to multiple regions and that each region uses a separate endpoint. What should you do?

A.Deploy a single Cognitive Services account and enable the multi-region feature.

B.Deploy a single Cognitive Services account and use the same endpoint for all regions.

C.Deploy multiple Cognitive Services accounts, one per region, and configure the application to use the appropriate endpoint.

D.Deploy a single Cognitive Services account and use Azure Traffic Manager to route requests to different regions.

AnswerC

Each Cognitive Services account is created in a specific region and provides a unique endpoint. By deploying one account per region, you ensure that each region has its own endpoint, enabling regional isolation and compliance. The application can select the endpoint based on the user's location or other logic.

Why this answer

To have separate endpoints per region, you must deploy a Cognitive Services account in each region. Each account provides a unique endpoint and key. This approach also allows for regional failover if needed.

The application must be designed to select the correct endpoint based on the region.

Exam trap

The trap here is assuming that a single Cognitive Services account can serve multiple regions or that a feature like multi-region exists.

Full explanation →

506

MCQhard

You are developing a generative AI application using Azure OpenAI. The application must log all prompts and completions for auditing, but sensitive data such as personal health information (PHI) must be redacted before logging. Which approach should you use?

A.Use Azure AI Language's PII detection to identify and redact PHI in prompts and completions before writing to your logging store.

B.Enable Azure OpenAI diagnostic settings to send logs to Azure Monitor, and rely on built-in PHI redaction.

C.Configure Azure OpenAI to not log prompts and completions, and instead log only token counts and model names.

D.Store logs in an Azure Storage account with encryption at rest and enable soft delete.

AnswerA

Azure AI Language provides PII detection that can identify and redact personal health information and other sensitive data. By calling this service on prompts and completions before logging, you ensure that only redacted text is stored. This meets auditing needs while protecting sensitive data.

Why this answer

To audit prompts and completions while protecting PHI, integrate Azure AI Language PII detection into your logging pipeline. Redact sensitive entities before persisting logs. This satisfies auditing requirements and compliance mandates, whereas relying on diagnostic logs or storage encryption alone would leave PHI exposed.

Exam trap

The trap here is assuming that Azure OpenAI diagnostic logs or storage encryption automatically redact PHI, when redaction must be performed explicitly before logging.

Full explanation →

507

MCQhard

Your Azure AI Search indexer is failing to index documents from an Azure Blob Storage container. The error message shows 'AccessDenied'. What is the most likely cause?

A.The blob container name is misspelled in the datasource.

B.The indexer execution interval is set too high.

C.The search service's managed identity lacks 'Storage Blob Data Reader' role.

D.The index schema does not match the blob metadata.

AnswerC

Blob indexers authenticate to storage using the search service's managed identity. Without the Storage Blob Data Reader role assignment on the container or account, the indexer cannot read blobs, producing the AccessDenied error described in the stem.

Why this answer

The 'AccessDenied' error indicates that the Azure AI Search service lacks the necessary permissions to read data from the Azure Blob Storage container. The most likely fix is to assign the 'Storage Blob Data Reader' role to the search service's system-assigned managed identity, which grants read access to blob containers and their contents.

Exam trap

Microsoft often tests the distinction between authentication (who you are) and authorization (what you can do), leading candidates to confuse a missing role assignment with a configuration error like a misspelled container name or schema mismatch.

How to eliminate wrong answers

Option A is wrong because a misspelled container name would cause a 'ContainerNotFound' or 'InvalidContainerName' error, not an 'AccessDenied' error. Option B is wrong because the indexer execution interval affects scheduling, not authentication or authorization; a high interval would simply delay indexing, not cause an access denial. Option D is wrong because schema mismatches between the index and blob metadata result in indexing failures with errors like 'FieldNotFound' or 'CannotConvertValue', not 'AccessDenied'.

Full explanation →

508

MCQhard

You have deployed a chatbot using Azure OpenAI with a system message as shown. The chatbot sometimes provides incorrect answers that are not supported by the sources. What is the most likely cause?

A.Content filters are incorrectly configured, allowing harmful content.

B.The data ingestion pipeline has errors, so the sources are not available.

C.The temperature parameter is too low, causing repetitive answers.

D.The system message does not guarantee grounding; the model may still hallucinate.

AnswerD

System messages steer tone and boundaries but cannot enforce factual grounding; the model still generates tokens probabilistically, so unsupported claims remain possible. Since the stem's constraint is answers not supported by sources, hallucination persists regardless of prompt wording. Retrieval augmentation or grounding data is required to constrain outputs to source content.

Why this answer

A system message in Azure OpenAI provides instructions and context but does not enforce factual grounding. The model can still generate responses that are not supported by the provided sources (hallucination), especially if the system message is not explicitly designed to restrict the model to only use the given data. Grounding requires additional techniques like retrieval-augmented generation (RAG) or explicit constraints in the prompt.

Exam trap

The trap here is that candidates often assume a system message is sufficient to enforce factual accuracy, confusing instruction-following with grounded generation, and overlook the need for retrieval-augmented generation or explicit source constraints.

How to eliminate wrong answers

Option A is wrong because content filters control the safety and appropriateness of outputs, not the factual accuracy or grounding of responses; they would not cause unsupported answers. Option B is wrong because if the data ingestion pipeline had errors making sources unavailable, the chatbot would likely fail to retrieve data or return errors, not produce plausible but incorrect answers. Option C is wrong because a low temperature parameter makes outputs more deterministic and repetitive, but it does not cause hallucination; in fact, lower temperature often reduces randomness and can improve consistency with training data, but it does not guarantee grounding to specific sources.

Full explanation →

509

MCQmedium

You are building a solution that must detect the sentiment polarity (positive, negative, neutral, or mixed) of incoming customer support tickets written in English and German. You want to use a single Azure AI Language resource and avoid maintaining separate models per language. Which approach should you use?

A.Call the Sentiment Analysis feature of Azure AI Language with the language parameter set to 'auto' so the service detects the language and returns sentiment for both English and German text.

B.Create two separate Azure AI Language resources, one configured for English and one for German, and route tickets based on a language detection step.

C.Train a custom sentiment model in Azure AI Language using labeled German and English tickets, then deploy it as a custom single-label classification project.

D.Use the Translator service to convert German tickets to English, then call Sentiment Analysis only on the translated English text.

AnswerA

Azure AI Language Sentiment Analysis supports automatic language detection when the language parameter is omitted or set to 'auto'. The service identifies the language and returns sentiment labels and confidence scores for English and German without requiring separate resources or custom models.

Why this answer

The prebuilt Sentiment Analysis capability in Azure AI Language natively supports multiple languages and can automatically detect the input language. Pointing a single resource at both English and German tickets avoids the overhead of separate resources, translation pipelines, or custom model training while still returning polarity and confidence scores.

Exam trap

The trap here is assuming that multilingual sentiment requires separate resources or a translation step, when Azure AI Language already detects language and analyzes sentiment natively.

Full explanation →

510

Multi-Selectmedium

Which TWO actions should you take to reduce the cost of using Azure OpenAI for a chatbot that handles high traffic?

Select 2 answers

A.Increase max_tokens to reduce the number of requests.

B.Use the batch API for non-real-time requests.

C.Implement caching for frequently asked questions.

D.Lower the temperature to 0 to reduce token usage.

E.Fine-tune the model to reduce prompt length.

AnswersB, C

The batch API processes asynchronous, non-real-time workloads at a 50% discount compared with standard global deployments, directly satisfying the stem's cost-reduction constraint for high-traffic chatbots. Offloading tolerant requests frees real-time capacity, lowering overall spend while preserving interactive latency for genuine user conversations.

Why this answer

Option B is correct because the Azure OpenAI Batch API processes asynchronous, non-real-time workloads at a 50% discount compared to standard global pricing, so routing any chatbot requests that don't need immediate responses through batch jobs directly lowers per-token cost. Option C is correct because caching responses to frequently asked questions (for example with Azure Cache for Redis or a semantic cache) means repeated identical or similar prompts are served without calling the model at all, eliminating token charges for that traffic and reducing overall request volume. Option A is not correct because increasing max_tokens raises the maximum output length and therefore can increase, not decrease, token consumption and cost.

Option D is not correct because temperature controls randomness in sampling, not the number of tokens generated, so setting it to 0 does not reduce token usage or cost. Option E is not correct because fine-tuning does not shorten the prompt by itself and adds training and hosting costs; prompt length is reduced through techniques like prompt compression or shorter system messages, not fine-tuning alone.

Exam trap

The trap here is that candidates confuse token-related parameters (max_tokens, temperature) with cost-saving mechanisms, when in reality cost reduction for high-traffic chatbots relies on architectural patterns like batching and caching, not on tweaking inference parameters.

Full explanation →

511

MCQmedium

You run the PowerShell script above to call the Text Analytics API. The response shows a sentiment label of 'positive' with a score of 0.99. However, you expected 'negative' because the word 'excellent' was meant to be sarcastic. What is the most likely reason for this result?

A.The score threshold for negative sentiment is too high.

B.The language parameter is set incorrectly.

C.The API version does not support sentiment analysis.

D.The model does not detect sarcasm and interprets the text literally.

AnswerD

Sentiment analysis models classify lexical polarity, not pragmatic intent, so "excellent" registers as strongly positive regardless of context. Sarcasm detection requires pragmatic inference that the Text Analytics sentiment model does not perform, satisfying the stem's constraint that the expected negative label never appears.

Why this answer

The Text Analytics API performs sentiment analysis using a machine learning model that evaluates the literal wording of the text. It does not have built-in capability to detect sarcasm, irony, or implied meaning. Therefore, the word 'excellent' is interpreted as positive regardless of the intended sarcastic tone, resulting in a high positive score.

Exam trap

The trap here is that candidates may assume the API can infer sarcasm or implied sentiment, when in fact the model performs only literal, surface-level sentiment analysis based on word choice and phrase patterns.

How to eliminate wrong answers

Option A is wrong because the score threshold for negative sentiment is not a configurable parameter in the Text Analytics API; the API returns a continuous sentiment score (0 to 1) and a label based on the model's confidence, not a threshold set by the user. Option B is wrong because the language parameter, if set incorrectly (e.g., 'en' for English), would cause the API to fail or return an error, not produce a positive sentiment for a sarcastic negative statement. Option C is wrong because the API version used in the script (v3.0 or later) fully supports sentiment analysis; the issue is not version-related but a model limitation.

Full explanation →

512

MCQeasy

You are deploying a conversational AI solution using Microsoft Copilot Studio. The solution must provide responses based on data from an internal knowledge base stored in SharePoint. Which feature should you configure?

A.Configure Azure AI Language with custom question answering.

B.Create an Azure Bot Service with a QnA Maker knowledge base.

C.Enable Generative Answers and add SharePoint as a data source.

D.Use Azure OpenAI Service with data integration (Bring Your Own Data).

AnswerC

Generative Answers retrieves and synthesises responses from configured enterprise data sources, and SharePoint is a natively supported source in Microsoft Copilot Studio. Adding it grounds the conversational agent in the internal knowledge base, satisfying the requirement to answer from SharePoint content rather than relying on the model's pretrained knowledge.

Why this answer

Microsoft Copilot Studio natively supports Generative Answers, which allows the AI to dynamically generate responses from specified data sources without requiring a separate QnA or custom question-answering service. By adding SharePoint as a data source, the solution can directly query the internal knowledge base stored in SharePoint and produce contextually relevant answers. This is the most straightforward and integrated approach within Copilot Studio for this scenario.

Exam trap

The trap here is that candidates often confuse the need for a separate QnA service (like Azure AI Language or QnA Maker) with Copilot Studio's built-in Generative Answers capability, not realizing that Copilot Studio can directly connect to SharePoint as a live data source without additional services.

How to eliminate wrong answers

Option A is wrong because Azure AI Language with custom question answering is a separate service that requires manual ingestion and management of Q&A pairs, whereas Copilot Studio's Generative Answers feature handles this automatically from SharePoint. Option B is wrong because Azure Bot Service with QnA Maker is a legacy approach that is now deprecated in favor of Copilot Studio and Generative Answers; QnA Maker also requires explicit Q&A pair extraction and does not natively integrate with SharePoint as a live data source. Option D is wrong because Azure OpenAI Service with data integration (Bring Your Own Data) is designed for custom GPT models and requires more complex setup and indexing, while Copilot Studio provides a simpler, out-of-the-box solution for connecting to SharePoint.

Full explanation →

513

MCQmedium

A company is building a chatbot using Azure OpenAI Service to handle customer inquiries. The bot sometimes responds with incorrect or fabricated information. The team wants to ground the model responses using their own product documentation stored in Azure Cognitive Search. Which configuration should they implement?

A.Use Azure AI Document Intelligence to extract text and generate embeddings, then store them in a vector database for direct similarity search.

B.Fine-tune the GPT model on the product documentation dataset.

C.Enable the semantic ranker in Azure Cognitive Search to improve the relevance of search results.

D.Configure Azure OpenAI on your data with Azure Cognitive Search as the data source.

AnswerD

Azure OpenAI on your data retrieves relevant chunks from Azure Cognitive Search and injects them into the prompt, grounding responses in your product documentation and reducing fabrication. This retrieval-augmented approach directly satisfies the requirement to constrain answers to indexed, authoritative content rather than the model's parametric memory.

Why this answer

It directly integrates Azure Cognitive Search as a data source for Azure OpenAI, enabling the model to retrieve and ground its responses in the indexed product documentation. This approach uses a 'retrieve-then-read' pattern where the search results are injected into the prompt, reducing hallucinations by constraining the model's output to verified content.

Exam trap

The trap here is that candidates confuse improving search relevance (semantic ranker) with the end-to-end grounding process, forgetting that the model must actually receive and be constrained by the retrieved data to prevent fabrication.

How to eliminate wrong answers

Option A is wrong because while Azure AI Document Intelligence can extract text and generate embeddings, storing them in a vector database for direct similarity search lacks the integrated grounding mechanism with Azure OpenAI; it would require custom orchestration to feed results into the model's context. Option B is wrong because fine-tuning the GPT model on product documentation would embed the data into the model's weights, which is costly, prone to overfitting, and does not allow real-time updates or retrieval of specific documents; it also does not guarantee factual accuracy for dynamic queries. Option C is wrong because enabling the semantic ranker in Azure Cognitive Search only improves search result relevance but does not connect the search results to Azure OpenAI for response generation; grounding requires the model to use the retrieved content, not just better search rankings.

Full explanation →

514

MCQeasy

You are building a generative AI solution using Azure OpenAI. The solution must generate responses that include factual information from a set of internal documents that are updated frequently. You need to ensure the model uses the latest document versions without retraining. What should you do?

A.Fine-tune the Azure OpenAI model each time the documents are updated.

B.Use Azure OpenAI with a retrieval-augmented generation pattern that queries an up-to-date search index of the documents.

C.Increase the model's temperature setting to encourage more creative and up-to-date responses.

D.Embed the documents directly into the model's prompt by concatenating all of them in each request.

AnswerB

RAG retrieves relevant passages from an index at query time, so when documents are updated and reindexed, the model automatically uses the latest content. This avoids retraining and ensures responses are grounded in current information. It is the standard approach for dynamic knowledge bases.

Why this answer

Retrieval-augmented generation with a search index allows the model to access current document content at inference time. When documents change, you update the index, and the model retrieves the latest passages. This avoids fine-tuning and ensures responses reflect the most recent information, which is critical for frequently updated knowledge bases.

Exam trap

The trap here is thinking that fine-tuning or higher temperature can keep a model current, when in fact retrieval from a refreshed index is what provides up-to-date grounding.

Full explanation →

515

MCQeasy

You are designing a generative AI assistant that must produce deterministic, repeatable answers for a compliance workflow. The assistant uses Azure OpenAI chat completions and must return the same output for identical prompts. Which parameter setting should you apply?

A.Set frequency_penalty to 2 and presence_penalty to 2.

B.Set temperature to 0 and top_p to 1.

C.Increase the max_tokens value and enable streaming responses.

D.Set temperature to 1 and top_p to 0.95.

AnswerB

Temperature 0 makes the model select the highest-probability token at each step, and top_p 1 disables nucleus sampling restrictions. Together they minimize randomness, producing nearly deterministic and repeatable outputs for identical prompts and parameters, which matches the compliance requirement for consistent answers.

Why this answer

Deterministic output requires minimizing sampling randomness. Temperature 0 selects the most probable token each step, and top_p 1 disables nucleus truncation, so identical inputs produce nearly identical outputs. Creative temperatures, penalties, or length and streaming controls do not provide repeatability and can introduce variation across runs.

Exam trap

The trap here is confusing output length or streaming controls with parameters that actually govern randomness.

Full explanation →

516

MCQmedium

You are deploying an Azure AI solution that uses Azure AI Search. The solution must be able to index data from an Azure SQL database and provide search results to an application. You need to ensure that the search service can access the database securely without storing credentials in the indexer definition. What should you do?

A.Configure the Azure AI Search service to use a managed identity and grant it access to the Azure SQL database.

B.Store the Azure SQL database connection string in Azure Key Vault and reference it in the indexer definition.

C.Use SQL authentication with a username and password, and encrypt the connection string using Azure AI Search's built-in encryption.

D.Configure the Azure SQL database to allow access from all Azure services and use a firewall rule.

AnswerA

By enabling a managed identity on the Azure AI Search service and granting that identity the appropriate permissions (such as db_datareader) on the Azure SQL database, you can avoid storing credentials in the indexer. The indexer can then use the managed identity to authenticate to the database, providing secure access without secrets.

Why this answer

Using a managed identity for the Azure AI Search service and granting it access to the Azure SQL database allows the indexer to authenticate without storing credentials. This is the most secure and recommended approach for service-to-service authentication in Azure.

Exam trap

The trap here is thinking that storing credentials in Key Vault eliminates the need for a managed identity, but the search service still requires an identity to access Key Vault, and the indexer would need to reference the secret, which is a form of credential storage.

Full explanation →

517

MCQhard

A logistics company runs an Azure AI Agent Service agent that must call an internal REST API to book delivery slots. The API requires a per-tenant OAuth token that must never be visible in run logs or agent definitions. The agent is invoked by many tenants through your own web application. Which approach should you use to give the agent access to the API securely?

A.Define the API as an OpenAPI tool with a project connection that holds the credential, and pass the tenant identity as a non-secret parameter.

B.Store the tenant token in the agent's instructions and let the model include it in the function call arguments.

C.Have the agent emit the tenant ID, then let your web application call the booking API directly and append the response as a user message.

D.Create a function tool that returns the tenant token to the model so the model can place it in the next function call.

AnswerA

OpenAPI tools let the service make the HTTP call on the agent's behalf, and the credential lives in the project connection rather than in messages or agent configuration. Passing only a tenant identifier as a tool parameter keeps the secret out of logs while still letting each invocation resolve the correct authorization behind the connection.

Why this answer

Using an OpenAPI tool with a project connection keeps credentials server-side while still letting the agent decide when to invoke the booking API. The tenant identifier travels as an ordinary parameter, so logs remain free of secrets and per-tenant authorization is resolved by the connection at call time rather than by the model.

Exam trap

The trap here is treating function tools as the only way to call external APIs, when OpenAPI tools with managed connections are designed precisely to keep credentials out of model-visible data.

Full explanation →

518

Multi-Selecteasy

Which TWO Azure services are used together to build a custom question-answering solution?

Select 2 answers

A.Azure AI Language (Custom Question Answering)

B.Azure AI Computer Vision

C.Azure AI Speech

D.Azure AI Search

E.Azure AI Translator

AnswersA, D

Azure AI Language's Custom Question Answering feature holds the question-and-answer knowledge base, project, and trained model that matches user queries to answers. It satisfies the scenario by supplying the language understanding and answer-generation component of the custom question-answering solution.

Why this answer

Azure AI Language (Custom Question Answering) is correct because it provides the natural-language processing layer that ingests FAQ documents, URLs, and structured sources to create a knowledge base and returns precise answers to user questions. Azure AI Search is correct because it is the underlying retrieval engine that indexes the knowledge base content and performs the semantic/keyword search that Custom Question Answering relies on to match questions to answers. Together they form the standard architecture for a custom question-answering solution: AI Language builds and manages the knowledge base, while AI Search stores and queries the indexed content.

Azure AI Computer Vision is for image analysis (OCR, object detection), Azure AI Speech handles speech-to-text/text-to-speech, and Azure AI Translator performs language translation, none of which are required components for building the question-answering knowledge base and retrieval pipeline.

Exam trap

The trap here is that candidates often assume Azure AI Speech or Azure AI Translator are needed for a 'custom' solution, but the core requirement is a searchable knowledge base, which is provided by Azure AI Search, not by speech or translation services.

Full explanation →

519

MCQeasy

You are developing a solution that uses Azure AI Translator to translate documents from English to French. You need to ensure that the translated text maintains the original formatting, such as HTML tags. Which feature should you use?

A.Use the Translator glossary to preserve formatting.

B.Use the Translator parallel corpus to train a custom model.

C.Set the content type to 'text/html' in the translation request.

D.Use the Transliterate method to preserve formatting.

AnswerC

Setting the content type to 'text/html' instructs Azure AI Translator to treat the input as markup, preserving HTML tags and their positions in the translated output. This directly satisfies the requirement that translated text maintains original formatting.

Why this answer

Azure AI Translator's Translate API accepts a contentType parameter, and setting it to 'text/html' tells the service to treat the input as HTML, preserving tags and only translating the human-readable text nodes. This is the documented mechanism for maintaining markup structure during translation. Without it, the service treats input as plain text and may mangle or translate tag names.

Exam trap

AI-102 often tests the confusion between translation customization features (glossary, custom models) and request-level parameters like contentType, so candidates must know that formatting preservation is a parameter setting, not a model-training feature.

How to eliminate wrong answers

Option A is wrong because a glossary customizes terminology translation, not markup preservation — it has no effect on HTML tags. Option B is wrong because a parallel corpus is used to train custom translation models for domain-specific phrasing, not to preserve formatting. Option D is wrong because Transliterate converts text between scripts (e.g., Latin to Cyrillic) without translating meaning and does not handle HTML structure.

Full explanation →

520

MCQeasy

Your company deploys an Azure AI Document Intelligence solution to extract data from invoices. During testing, you notice that some fields are not being extracted correctly, especially for invoices from a specific vendor with a non-standard layout. You need to improve extraction accuracy for this vendor's invoices. What should you do?

A.Enable OCR on the documents and use regular expressions to extract fields.

B.Convert the invoices to a standard format before processing.

C.Train a custom model using labeled samples of the vendor's invoices.

D.Use the prebuilt invoice model with confidence threshold adjustment.

AnswerC

A custom model trained on labelled samples of that vendor's invoices learns its non-standard layout and field positions, which the prebuilt invoice model cannot capture. This directly targets the extraction accuracy problem for that specific vendor.

Why this answer

Training a custom model using labeled samples of the vendor's invoices allows Document Intelligence to learn the non-standard layout, improving extraction accuracy. Option A is incorrect because while OCR and regex can extract text, they are not effective for structured data extraction from variable layouts. Option B is incorrect because converting invoices to a standard format is time-consuming and may lose important data; it's better to train a model on the actual invoices.

Option D is incorrect because the prebuilt invoice model is designed for standard invoice layouts; adjusting confidence thresholds won't fix extraction accuracy for non-standard formats.

Full explanation →

521

MCQhard

A company uses Azure AI Search to index a large collection of scanned invoices stored in Azure Blob Storage. They have a skillset that includes an OCR skill to extract text from the invoices. The indexer is configured to run every night. They notice that the indexer takes a long time to complete and sometimes times out. They want to optimize the indexer performance without reducing the quality of the extracted text. What should they do?

A.Switch from the OCR skill to the Text Merge skill to combine text from multiple pages.

B.Enable incremental enrichment and cache the enriched data to avoid reprocessing unchanged documents.

C.Increase the indexer's batch size and reduce the number of parallel indexers.

D.Reduce the OCR skill's image resolution by setting the 'imageAction' to 'none' and relying on the document's embedded text.

AnswerB

This is correct because incremental enrichment with caching allows the indexer to skip documents that have not changed since the last run, reusing previously enriched data. This reduces the amount of OCR processing required, significantly improving performance for subsequent runs. It maintains text quality because unchanged documents are not reprocessed, and only new or modified documents are enriched.

Why this answer

Enabling incremental enrichment and caching is the best approach to optimize indexer performance for recurring runs. It avoids reprocessing documents that have not changed, thus reducing the workload on the OCR skill and other enrichments. This maintains the quality of extracted text because only new or modified documents are processed, while unchanged documents reuse cached enrichments.

Other options either reduce quality or do not address the performance bottleneck.

Exam trap

The trap here is thinking that reducing image resolution or disabling OCR will speed up the indexer, but that sacrifices the required text extraction quality.

Full explanation →

522

MCQhard

You are designing a solution that uses Azure AI Language custom question answering. The knowledge base contains 200 question-answer pairs. Users report that the bot sometimes returns answers to questions that are semantically similar but not actually asked. You need to reduce these false positives while maintaining the ability to answer paraphrased questions. What should you configure?

A.Set the 'default answer' to a custom message indicating no answer was found.

B.Enable the 'Enable active learning' option.

C.Add more alternative questions to each question-answer pair.

D.Increase the confidence threshold score for the project.

AnswerD

Raising the confidence threshold makes the bot return an answer only when the match score exceeds a higher value. This reduces false positives from semantically similar but incorrect matches. It still allows paraphrased questions to be answered if their score is high enough. The threshold is adjustable in the project settings and directly controls the trade-off between precision and recall.

Why this answer

The confidence threshold determines the minimum score required for an answer to be returned. By increasing it, the bot becomes more selective, reducing false positives from semantically similar questions. Active learning and alternative questions improve coverage but do not directly filter low-confidence matches.

The default answer is only a fallback and does not affect matching behavior.

Exam trap

The trap here is confusing active learning with runtime filtering; active learning improves the model over time but does not change the immediate matching threshold.

Full explanation →

523

MCQeasy

Refer to the exhibit. You are deploying a GPT-4 model using Azure OpenAI Service. The deployment uses the Standard scale type. Which statement is true about this deployment?

A.The model version is not specified and will default to the latest.

B.Content filtering is disabled for this deployment.

C.The deployment uses provisioned throughput with reserved capacity.

D.The deployment uses pay-as-you-go pricing and global rate limits.

AnswerD

Standard deployments bill per token on a pay-as-you-go basis and draw on the shared, globally pooled quota for that model, so throughput varies with regional demand. Provisioned scale types instead reserve dedicated capacity with predictable latency and a fixed hourly charge.

Why this answer

The Standard scale type in Azure OpenAI Service uses pay-as-you-go pricing, where you are billed based on the number of tokens processed. It also enforces global rate limits (e.g., tokens per minute) that apply across all deployments in the region, rather than providing dedicated capacity. Option D correctly identifies these characteristics.

Exam trap

Microsoft often tests the distinction between Standard (pay-as-you-go with rate limits) and Provisioned (reserved capacity) scale types, and candidates mistakenly associate 'Standard' with default model versions or disabled content filtering.

How to eliminate wrong answers

Option A is wrong because when deploying a GPT-4 model, the model version must be explicitly specified; it does not default to the latest version. Option B is wrong because content filtering is enabled by default for all Azure OpenAI Service deployments and cannot be disabled through the scale type setting. Option C is wrong because provisioned throughput with reserved capacity is a feature of the Provisioned scale type, not the Standard scale type.

Full explanation →

524

MCQeasy

A company wants to use Azure AI services to extract text from scanned PDF documents. Which Azure AI service should they use?

A.Azure AI Language Understanding (LUIS)

B.Azure AI Document Intelligence

C.Azure AI Computer Vision API

D.Azure Cognitive Search

AnswerB

Scanned PDFs contain images, not embedded text, so optical character recognition is required. Azure AI Document Intelligence's prebuilt read model performs OCR and layout extraction, returning text and structure from image-only documents, which satisfies the extraction requirement.

Why this answer

Azure AI Document Intelligence (formerly Form Recognizer) is the correct service because it is specifically designed for extracting text, tables, and key-value pairs from scanned PDFs and images using optical character recognition (OCR) and deep learning models. Unlike general OCR APIs, Document Intelligence can handle complex layouts and preserve document structure, making it ideal for this use case.

Exam trap

The trap here is that candidates often confuse the general-purpose Computer Vision OCR API with the specialized Document Intelligence service, overlooking that Document Intelligence offers superior layout understanding and prebuilt models for document-centric extraction tasks.

How to eliminate wrong answers

Option A is wrong because Azure AI Language Understanding (LUIS) is a conversational language understanding service for intent and entity extraction from natural language utterances, not for extracting text from scanned documents. Option C is wrong because while Azure AI Computer Vision API includes OCR capabilities, it is a general-purpose image analysis service that lacks the specialized layout analysis, table extraction, and form understanding features that Document Intelligence provides for scanned PDFs. Option D is wrong because Azure Cognitive Search is a search indexing and query service that can index extracted text but does not perform the initial extraction from scanned PDFs itself.

Full explanation →

525

MCQeasy

You are designing a conversational AI solution using Microsoft Copilot Studio. The exhibit shows part of a topic configuration. What is the purpose of the 'triggers' section?

A.To define the response the bot sends

B.To define the authentication method

C.To define the conditions that activate the topic

D.To define the entities to extract

AnswerC

Triggers define the phrases or conditions that cause a topic to run during a conversation. They act as the entry point, matching user utterances so Copilot Studio routes the dialogue into the correct topic before its actions execute.

Why this answer

In Microsoft Copilot Studio, the 'triggers' section defines the conditions (such as specific phrases, intents, or keywords) that activate a topic. This is the entry point for the conversation flow, ensuring the bot responds appropriately when a user's input matches the defined trigger phrases. Option C correctly identifies this purpose.

Exam trap

The trap here is that candidates often confuse 'triggers' with 'responses' or 'entities', because triggers initiate the topic flow, but they do not define what the bot says or extract data—those are separate configuration elements within the topic canvas.

How to eliminate wrong answers

Option A is wrong because defining the response the bot sends is handled by the 'Message' or 'Question' nodes within the topic, not the triggers section. Option B is wrong because authentication methods are configured in the bot's settings or channel configuration, not within individual topic triggers. Option D is wrong because entities are extracted using the 'Entity recognition' feature or prebuilt entities, not defined in the triggers section; triggers focus on activation conditions, not data extraction.

Full explanation →

Microsoft Azure AI Engineer Associate AI-102 (AI-102) — Questions 451–525