Microsoft Azure AI Engineer Associate AI-102 AI-102 Questions 526–600 | Page 8/11

526

MCQmedium

A company uses Azure AI Vision to extract text from scanned invoices. They need to preserve the layout information, such as tables and key-value pairs, to automate data entry. Which Azure service should they use?

A.Azure AI Face API

B.Azure AI Custom Vision

C.Azure AI Vision Read API

D.Azure AI Document Intelligence (formerly Form Recognizer)

AnswerD

Document Intelligence is designed to extract text, tables, and key-value pairs from documents such as invoices. It provides prebuilt models for invoices that understand layout and can return structured data like invoice ID, date, and line items. This directly meets the need to preserve layout information and automate data entry, making it the correct choice for this scenario.

Why this answer

Document Intelligence provides prebuilt invoice models that extract text, tables, and key-value pairs while preserving layout. The Read API only returns text lines, and Custom Vision and Face API are for different domains. For automating data entry from invoices with tables, Document Intelligence is the appropriate service.

Exam trap

The trap here is assuming the Read API is sufficient for invoices, but it lacks layout understanding and structured field extraction that Document Intelligence provides.

Full explanation →

527

Multi-Selecthard

Your organization uses Azure AI Language to perform sentiment analysis and opinion mining on product reviews. You notice that the sentiment scores are often neutral even when the review text contains clearly positive or negative opinions. You suspect the model is not capturing the nuances. Which three actions could improve the sentiment analysis accuracy? (Choose three.)

Select 3 answers

A.Provide more labeled training examples that cover a wider variety of writing styles and sentiments.

B.Pre-process the text with key phrase extraction to highlight important terms before sentiment analysis.

C.Use the opinion mining feature to capture sentiment targets and associated opinions.

D.Use the basic sentiment analysis API without any customization.

E.Enable the domain-specific model for 'Reviews' if available.

AnswersA, C, E

More diverse training data helps the model generalize better and capture nuances.

Why this answer

Providing more labeled training examples that cover a wider variety of writing styles and sentiments directly improves the custom model's ability to learn nuanced patterns. Azure AI Language's custom sentiment analysis relies on supervised learning; more diverse, high-quality labeled data helps the model generalize better and reduces the tendency to default to neutral scores for ambiguous or complex reviews.

Exam trap

The trap here is that candidates may assume pre-processing with key phrase extraction (option B) or using the basic API (option D) can fix model accuracy issues, when in fact only custom training (Azure AI Language custom sentiment analysis), opinion mining, or domain-specific models address the root cause of neutral scores due to lack of nuance.

Full explanation →

528

MCQeasy

You are deploying an Azure OpenAI model for a public-facing FAQ assistant. The assistant must answer only questions covered by a fixed set of approved topics, and any off-topic question should receive a polite refusal. Which approach most directly enforces this behavior?

A.Raise the frequency_penalty parameter to reduce repeated off-topic phrases.

B.Deploy the model with the lowest available quota tier to limit how many questions users can ask.

C.Set the model deployment's content filter to block the 'violence' category at high severity.

D.Configure a system message that defines the allowed topics and instructs the model to refuse anything outside them.

AnswerD

The system message sets the model's operating instructions and persona for every turn in the conversation. Defining the permitted topics and the refusal behavior there gives the model a consistent boundary to apply across all user turns. This is the most direct, low-overhead way to constrain scope for a simple FAQ assistant.

Why this answer

The system message is the model's persistent instruction set, making it the natural place to declare allowed topics and refusal behavior. For a fixed FAQ scope, this directly shapes every response. The other settings influence sampling, safety categories, or capacity, none of which restrict the assistant to approved topics.

Exam trap

The trap here is confusing safety content filters with topical scope control, when filters only block harmful categories rather than off-topic questions.

Full explanation →

529

MCQeasy

You plan to use Azure AI Content Safety to detect hate speech in user-generated content. Which type of content safety is most appropriate for this scenario?

A.Custom categories

B.Image moderation

C.Prompt Shields

D.Text moderation

AnswerD

Text moderation analyses written user-generated content and returns severity scores for hate, violence, self-harm and sexual categories. This satisfies the stem's requirement to detect hate speech in text, unlike image moderation, which only classifies visual content.

Why this answer

Text moderation is the correct choice because Azure AI Content Safety's text moderation API is specifically designed to detect and filter hate speech, along with other harmful content categories like violence and self-harm, in user-generated text. It uses machine learning classifiers trained on a vast corpus to assign severity scores across predefined categories, making it the direct and most appropriate tool for this scenario.

Exam trap

The trap here is that candidates may confuse the broad 'text moderation' capability with the more specialized 'Prompt Shields' feature, mistakenly thinking prompt injection protection is the same as hate speech detection, or assume 'custom categories' are needed when the built-in hate category already suffices.

How to eliminate wrong answers

Option A is wrong because custom categories allow you to define your own specific terms or patterns for blocking, but they are not the primary or most appropriate method for detecting broad, nuanced hate speech; the service's built-in text moderation categories already cover hate speech comprehensively. Option B is wrong because image moderation is designed to analyze visual content for adult, racy, or violent imagery, not to detect hate speech in text. Option C is wrong because Prompt Shields are a feature of Azure AI Content Safety that protects against prompt injection attacks in generative AI applications, not for detecting hate speech in general user-generated content.

Full explanation →

530

MCQhard

You are creating a new Custom Vision project with the above JSON. The domainId corresponds to the 'Logo' domain. Which type of model will this project train?

A.An object detection model for logo detection

B.An optical character recognition model

C.A multilabel image classification model for logo detection

D.A general image classification model

AnswerC

The Logo domain is a multilabel classification domain, meaning each image can be assigned multiple logo tags simultaneously rather than one exclusive label. Training with that domainId therefore produces a multilabel image classification model for logo detection.

Why this answer

The 'Logo' domain in Custom Vision is specifically designed for image classification tasks, not object detection. When you create a project with the 'Logo' domain, it trains a multilabel image classification model, meaning each image can be assigned multiple labels (e.g., multiple logos in one image). This domain is optimized for identifying and classifying logos within images, making it distinct from object detection or general classification.

Exam trap

The trap here is that candidates often confuse the 'Logo' domain with object detection, assuming it draws bounding boxes around logos, when in fact it performs multilabel classification without localization.

How to eliminate wrong answers

Option A is wrong because the 'Logo' domain does not correspond to object detection; object detection requires a domain like 'General (Object Detection)' or 'Logo (Object Detection)' if available, and the JSON specifies the 'Logo' domain which is for classification. Option B is wrong because optical character recognition (OCR) is not a Custom Vision domain; OCR is handled by Azure Cognitive Services like Computer Vision's Read API, not Custom Vision. Option D is wrong because while the 'Logo' domain is a type of image classification, it is specifically a multilabel classification model, not a general image classification model (which typically uses single-label classification).

Full explanation →

531

MCQmedium

You are deploying a generative AI application using Azure OpenAI Service. The application must generate responses in multiple languages while maintaining high accuracy. You need to minimize token usage. Which approach should you recommend?

A.Use a base model with a system message to output in the desired language

B.Translate all input to English and then translate output back

C.Fine-tune a model for each target language

D.Use a separate deployment for each language

AnswerA

A system message steers the base model to respond in the target language, avoiding separate per-language models or lengthy translation prompts. This keeps prompt tokens minimal, satisfying the requirement to minimise token usage while retaining accuracy.

Why this answer

Azure OpenAI Service base models (e.g., GPT-4) natively support multilingual generation via a system message that sets the desired output language. This approach avoids the overhead of translation pipelines or fine-tuning, directly minimizing token usage while maintaining high accuracy through the model's inherent multilingual capabilities.

Exam trap

Azure AI-102 often tests the misconception that translation pipelines or fine-tuning are necessary for multilingual support, when in fact a system message on a base model achieves the goal with lower token usage and complexity.

How to eliminate wrong answers

Option B is wrong because translating all input to English and then back adds significant token overhead (doubling input/output tokens) and introduces translation errors that degrade accuracy, contradicting the requirement to minimize token usage. Option C is wrong because fine-tuning a separate model for each target language is resource-intensive, requires large labeled datasets per language, and does not reduce token usage compared to a single base model with a system message. Option D is wrong because using a separate deployment for each language multiplies infrastructure costs and management complexity without any token savings, as each deployment still processes tokens for the same base model.

Full explanation →

532

MCQmedium

A company is using Azure Form Recognizer to extract data from invoices. The prebuilt model does not correctly extract a custom field that is specific to the company's invoices. What is the most appropriate action to improve extraction accuracy for this field?

A.Use the prebuilt model with a custom field mapping.

B.Train a custom model using labeled invoices that include the custom field.

C.Adjust the confidence threshold for the prebuilt model.

D.Retrain the prebuilt model with additional invoices.

AnswerB

A custom model trained on labelled invoices teaches Form Recognizer the layout and semantics of the company-specific field, which the prebuilt invoice model cannot infer. Labelled samples supply the field's position and value patterns, directly improving extraction accuracy for that field.

Why this answer

The prebuilt Form Recognizer model is designed for common invoice layouts and may not recognize company-specific fields. Training a custom model with labeled invoices that include the custom field allows the model to learn the field's location and semantics, significantly improving extraction accuracy for that specific field.

Exam trap

The trap here is that candidates may think prebuilt models can be customized via mapping or retraining, but Azure Form Recognizer prebuilt models are immutable and only custom models can be trained to recognize new fields.

How to eliminate wrong answers

Option A is wrong because prebuilt models do not support custom field mapping; they extract only predefined fields based on their training data. Option C is wrong because adjusting the confidence threshold only filters results based on confidence scores, it does not teach the model to recognize a new field. Option D is wrong because prebuilt models cannot be retrained; they are fixed by Microsoft and only custom models can be trained with additional data.

Full explanation →

533

MCQmedium

A healthcare organization uses Azure AI Language to extract medical entities from clinical notes. The solution must comply with HIPAA and data residency requirements. Which configuration is essential?

A.Enable diagnostic logging for all operations.

B.Use a customer-managed key (CMK) for encryption.

C.Enable private endpoint for the AI resource.

D.Create the AI resource in the required Azure region.

AnswerD

Data residency requires the Azure AI resource to reside in the mandated geography, because Azure AI Language processes and stores data in the resource's region. Creating the resource in the required region satisfies the residency constraint; HIPAA compliance is then addressed through the resulting regional deployment.

Why this answer

Data residency requirements dictate that the Azure AI Language resource must be physically located in the specific Azure region where the clinical notes and extracted medical entities are permitted to reside. Creating the resource in the required Azure region ensures that all data at rest and in transit stays within that geographic boundary, which is a fundamental compliance step for HIPAA and data residency. Other configurations like encryption keys or private endpoints enhance security but do not satisfy the core residency requirement.

Exam trap

The trap here is that candidates often confuse network-level security (private endpoints) or encryption controls (CMK) with data residency, assuming any security measure automatically satisfies geographic compliance requirements.

How to eliminate wrong answers

Option A is wrong because enabling diagnostic logging captures operational telemetry but does not enforce data residency or HIPAA compliance; it may even introduce additional data handling concerns. Option B is wrong because using a customer-managed key (CMK) controls encryption keys but does not control where the data is stored or processed; data residency is a separate requirement. Option C is wrong because enabling a private endpoint restricts network access to the AI resource via a VNet but does not change the physical region where the resource and its data reside.

Full explanation →

534

MCQhard

Refer to the exhibit. You are configuring a system message for an Azure OpenAI deployment. The assistant is still generating harmful code despite the instruction. Which additional measure should you implement?

A.Fine-tune the model on safe code examples.

B.Lower the temperature parameter to 0.

C.Add more examples to the prompt.

D.Enable Azure AI Content Safety with a custom blocklist for harmful code.

AnswerD

Prompt instructions alone cannot reliably block harmful code generation. Azure AI Content Safety filters model inputs and outputs against configurable harm categories, and a custom blocklist adds scenario-specific terms, enforcing the constraint at the platform layer regardless of what the system message says.

Why this answer

Azure AI Content Safety provides a dedicated content filtering layer that can block harmful code generation at the inference level, regardless of the system message. A custom blocklist allows you to define specific patterns (e.g., code snippets for malware) that the model is prohibited from outputting, enforcing safety beyond prompt instructions.

Exam trap

The trap here is that candidates often assume prompt engineering (system messages or few-shot examples) is sufficient for safety, but Azure OpenAI requires explicit content filtering via Azure AI Content Safety to reliably block harmful outputs at scale.

How to eliminate wrong answers

Option A is wrong because fine-tuning on safe code examples would require retraining the model, which is costly, time-consuming, and not a quick mitigation for an existing deployment; it also does not guarantee blocking of harmful code at inference time. Option B is wrong because lowering the temperature to 0 makes the model more deterministic but does not prevent it from generating harmful code if that code is the most likely completion. Option C is wrong because adding more examples to the prompt (few-shot prompting) can guide behavior but is unreliable for safety enforcement, as the model may still generate harmful code if the examples are not exhaustive or if the model overfits to the instruction.

Full explanation →

535

MCQeasy

You need to transcribe customer service calls into text for analysis. Which Azure service should you use?

A.Conversational Language Understanding

B.Azure AI Speech-to-Text

C.Azure AI Translator

D.Azure AI Text-to-Speech

AnswerB

Azure AI Speech-to-Text converts spoken audio into written transcripts, directly satisfying the requirement to transcribe customer service calls. Its real-time and batch transcription APIs handle telephony audio formats, enabling downstream text analysis without manual effort.

Why this answer

Azure AI Speech-to-Text (B) is the correct service for transcribing audio recordings of customer service calls into text. It provides real-time or batch transcription of spoken language, which is exactly what the scenario requires for subsequent analysis. The other services handle different tasks: understanding intent, translating text, or generating speech.

Exam trap

The trap here is that candidates confuse 'understanding language' (CLU) with 'transcribing speech' (Speech-to-Text), assuming CLU can directly process audio, when in reality CLU only works on text input.

How to eliminate wrong answers

Option A is wrong because Conversational Language Understanding (CLU) is designed to extract intents and entities from text, not to transcribe audio into text; it requires pre-transcribed input. Option C is wrong because Azure AI Translator translates text between languages, not from speech to text; it cannot process audio files. Option D is wrong because Azure AI Text-to-Speech converts text into spoken audio, the reverse of the required transcription workflow.

Full explanation →

536

MCQmedium

A company is building a chatbot using Azure Bot Service and Language Understanding (LUIS). The chatbot needs to handle user intents for booking flights and checking flight status. After testing, the chatbot frequently fails to distinguish between the two intents when users mention flight numbers. Which action should the engineer take to improve intent recognition?

A.Increase the number of intents to split the flight-related queries further.

B.Add more utterances that include flight numbers to the training data for both intents.

C.Reduce the confidence score threshold for intent detection.

D.Use the prebuilt domain for flight booking to improve accuracy.

AnswerB

Adding utterances containing flight numbers to both intents gives the model contrasting examples of the same entity in different contexts, letting LUIS learn which surrounding phrasing signals booking versus status checking. This directly addresses the stem's constraint: overlapping flight-number mentions that currently cause misclassification between the two intents.

Why this answer

Adding more utterances that include flight numbers to both intents provides LUIS with more varied examples of how flight numbers appear in natural language, enabling the model to learn distinguishing patterns. Without sufficient training data containing flight numbers, LUIS cannot reliably differentiate between 'BookFlight' and 'CheckFlightStatus' when users mention flight numbers, as the entity alone does not determine intent.

Exam trap

The trap here is that candidates often think reducing the confidence threshold or adding more intents will fix misclassification, but the real issue is insufficient representative training data for the specific ambiguous patterns (flight numbers) that cause confusion.

How to eliminate wrong answers

Option A is wrong because increasing the number of intents would further fragment the training data, making it harder for LUIS to distinguish between similar queries, and does not address the core issue of insufficient examples with flight numbers. Option C is wrong because reducing the confidence score threshold would cause more false positives, increasing misclassification rather than improving accuracy. Option D is wrong because prebuilt domains provide generic intents and entities that may not match the company's specific flight-related queries, and they do not solve the problem of distinguishing between two custom intents when flight numbers are present.

Full explanation →

537

MCQmedium

You are building a computer vision solution to detect defects on a manufacturing assembly line. The solution must process images in real-time with low latency, and you need to choose an Azure service. Which service should you use?

A.Azure Computer Vision API

B.Azure Video Indexer

C.Azure Form Recognizer

D.Azure Custom Vision

AnswerD

Azure Custom Vision trains and hosts a purpose-built image classification or object detection model, letting you deploy a compact domain-specific model to a prediction endpoint for low-latency inline defect scoring, satisfying the real-time assembly-line constraint.

Why this answer

Azure Custom Vision is the correct choice because it allows you to train a custom image classification or object detection model tailored to detect specific manufacturing defects. It supports real-time, low-latency inference via a Docker container deployed to edge devices or directly through the prediction API, meeting the assembly line's performance requirements.

Exam trap

The trap here is that candidates often choose Azure Computer Vision API (Option A) because it sounds like a general-purpose vision service, but they overlook the requirement for custom defect detection, which necessitates a trainable model like Custom Vision.

How to eliminate wrong answers

Option A is wrong because Azure Computer Vision API provides pre-trained models for general image analysis (e.g., OCR, tagging) and cannot be customized to detect specific manufacturing defects without retraining. Option B is wrong because Azure Video Indexer is designed for analyzing video content (e.g., speech, faces, scenes) and is not optimized for real-time, low-latency image processing on a per-frame basis. Option C is wrong because Azure Form Recognizer is specialized for extracting text and structure from documents (e.g., invoices, forms), not for detecting visual defects in manufacturing images.

Full explanation →

538

Multi-Selecthard

Which THREE factors should you consider when selecting a pricing tier for Azure Computer Vision in a production environment?

Select 3 answers

A.Availability of free tier

B.Type of storage account for images

C.Data residency requirements

D.Latency requirements

E.Transactions per second limit

AnswersC, D, E

May require specific region and tier.

Why this answer

Data residency requirements (Option C) are critical when selecting a pricing tier for Azure Computer Vision because the service processes images in specific regional data centers, and some tiers (e.g., Standard S0) support multi-region processing while others may be restricted. Compliance with regulations like GDPR or HIPAA may require that image data never leaves a particular geography, directly influencing which tier and region you can choose.

Exam trap

The trap here is that candidates often confuse the free tier's availability as a valid production option, or mistakenly think storage account type influences pricing tier selection, when in reality the key factors are operational constraints like TPS, latency, and data residency compliance.

Full explanation →

539

MCQeasy

A company wants to extract key-value pairs from scanned invoices using Azure AI. Which service should they use?

A.Read API

B.Custom Vision

C.OCR API

D.Azure AI Document Intelligence

AnswerD

Azure AI Document Intelligence's prebuilt invoice model extracts key-value pairs such as invoice number, date, and total directly from scanned documents, satisfying the requirement for structured field extraction. Its OCR and layout analysis handle scanned images, unlike generic vision or language services that lack invoice-specific field mapping.

Why this answer

Azure AI Document Intelligence (formerly Form Recognizer) is the correct choice because it is specifically designed to extract key-value pairs, tables, and structured data from scanned documents like invoices. Unlike the Read API or OCR API, which only return raw text or OCR output, Document Intelligence uses prebuilt models (e.g., 'prebuilt-invoice') that understand the semantic layout of invoices, enabling direct extraction of fields such as invoice number, date, and total amount.

Exam trap

The trap here is that candidates often confuse the Read API or OCR API with Document Intelligence because all three involve text extraction, but only Document Intelligence provides key-value pair extraction and document understanding capabilities.

How to eliminate wrong answers

Option A is wrong because the Read API extracts printed and handwritten text as lines and words, but it does not parse key-value pairs or understand document structure. Option B is wrong because Custom Vision is an image classification and object detection service, not designed for text extraction or document understanding. Option C is wrong because the OCR API (part of Computer Vision) performs optical character recognition to return raw text and bounding boxes, but it lacks the ability to identify and extract key-value pairs or structured fields from invoices.

Full explanation →

540

MCQeasy

A company wants to generate product descriptions for thousands of items using an Azure OpenAI GPT-4 model. They need to ensure the descriptions match a consistent brand voice. Which approach is most efficient and cost-effective?

A.Write a separate prompt for each product category

B.Use Azure OpenAI on your data with a vector database of brand guidelines

C.Set a system message with brand voice guidelines and use few-shot examples

D.Fine-tune a base model on existing product descriptions

AnswerC

A system message plus few-shot examples conditions one GPT-4 deployment to produce consistent brand-voice output across thousands of items, avoiding the cost and latency of fine-tuning or per-item prompt engineering. This satisfies the consistency and cost-efficiency constraints simultaneously.

Why this answer

Setting a system message with brand voice guidelines and providing few-shot examples allows the GPT-4 model to consistently apply the desired tone and style across all product descriptions without retraining. This approach is efficient and cost-effective as it avoids the high compute and data preparation costs of fine-tuning, while still enabling precise control over output through in-context learning.

Exam trap

Microsoft often tests the misconception that fine-tuning is always the best approach for consistency, but in Azure OpenAI, in-context learning via system messages and few-shot examples is more efficient and cost-effective for tasks like brand voice adherence, as fine-tuning is reserved for deep customization of model behavior.

How to eliminate wrong answers

Option A is wrong because writing a separate prompt for each product category would be highly inefficient and inconsistent, as it requires manual effort for thousands of items and does not leverage the model's ability to generalize from a single system message. Option B is wrong because using Azure OpenAI on your data with a vector database of brand guidelines is overkill for this task; vector databases are designed for retrieval-augmented generation (RAG) to ground responses in external data, but brand voice guidelines are better conveyed via system messages and examples, not as searchable documents. Option D is wrong because fine-tuning a base model on existing product descriptions is costly, requires significant labeled data, and risks overfitting to the training set, whereas in-context learning with a system message and few-shot examples achieves the same goal with far less expense and complexity.

Full explanation →

541

MCQmedium

You are building a multilingual customer support chatbot using Azure AI Language. The bot must understand user intents in English, Spanish, and French. You have pre-existing labeled data in English only. The solution should minimize manual labeling effort. Which approach should you recommend?

A.Use Azure AI Translator to detect the language and route to a rules-based intent handler for each language.

B.Build a separate CLU project for each language and use the English data to bootstrap labeling with active learning.

C.Use the multilingual option in conversational language understanding (CLU) and train on the English data only.

D.Translate the English labeled data into Spanish and French using Azure AI Translator, then train a separate CLU model per language.

AnswerC

CLU's multilingual option trains a single project across several languages, so English-labelled utterances transfer to Spanish and French inference. This satisfies the requirement to minimise manual labelling while still understanding intents in all three languages.

Why this answer

Azure AI Language's conversational language understanding (CLU) supports a multilingual project option that allows you to train a single model on labeled data in one language (e.g., English) and have it generalize to understand intents in other languages (e.g., Spanish and French) without needing additional labeled data. This directly minimizes manual labeling effort while still leveraging the pre-existing English data.

Exam trap

The trap here is that candidates often assume you must have labeled data in each target language or use translation, overlooking Azure's built-in multilingual support that enables zero-shot cross-lingual intent recognition.

How to eliminate wrong answers

Option A is wrong because it relies on a rules-based intent handler, which is not a natural language understanding approach and would require manual creation of rules for each language, defeating the goal of minimizing effort. Option B is wrong because building separate CLU projects for each language and using active learning still requires manual labeling effort for each language, as active learning only reduces but does not eliminate the need for labeled data in each target language. Option D is wrong because translating labeled data and training separate models per language introduces translation errors and doubles the training effort, which is more labor-intensive than using the multilingual CLU option.

Full explanation →

542

MCQmedium

You are implementing a knowledge mining solution using Azure AI Search with a custom skillset. The custom skill is an Azure Function that enriches documents with additional metadata. You need to ensure that the custom skill receives the entire document content as input. How should you configure the skill's context and inputs?

A.Set context to '/document/content' and input source to '/document/metadata'.

B.Set context to '/document/content' and input source to '/document/content'.

C.Set context to '/document' and input source to '/document/normalized_images/*'.

D.Set context to '/document' and input source to '/document/content'.

AnswerD

Setting context to `/document` makes the skill execute once per document rather than per page or chunk, while the input source `/document/content` passes the full extracted text of that document into the Azure Function. This satisfies the requirement that the custom skill receives the entire document content.

Why this answer

To pass the entire document content to a custom skill, the skill's context must be set to '/document' (the root of each document in the enrichment tree) and the input source must be '/document/content'. Setting context to '/document' ensures the skill executes once per document, and mapping the input to '/document/content' delivers the full text content to the Azure Function. This is the correct configuration for document-level enrichment.

Exam trap

AI-102 often tests the distinction between skill context (execution granularity) and input source (data passed to the skill), tricking candidates who set context to '/document/content' instead of '/document' or who confuse content with normalized_images.

How to eliminate wrong answers

Option A is wrong because while context '/document/content' would scope the skill to the content node, the input source '/document/metadata' would pass metadata rather than the full content, and the context is not the correct document-level scope for whole-document enrichment. Option B is wrong because setting context to '/document/content' scopes the skill to the content field itself rather than the document root, which can cause issues when the skill needs document-level context and multiple inputs. Option C is wrong because context '/document' is correct, but the input source '/document/normalized_images/*' passes normalized images (used for OCR/image skills), not the document's text content.

Full explanation →

543

Multi-Selecthard

A company uses Azure AI Speech to provide real-time transcription for customer support calls. The solution must handle multiple languages and filter out profanity. Which THREE configurations are needed?

Select 3 answers

A.Use the Batch Transcription REST API

B.Deploy a custom speech model for each language

C.Set the SpeechConfig.SpeechRecognitionLanguage property

D.Use the Speech SDK with intermediate results

E.Enable the ProfanityFilter option in the Speech SDK

AnswersC, D, E

Setting SpeechConfig.SpeechRecognitionLanguage specifies the spoken language the recogniser expects, satisfying the stem's multiple-language requirement. Without it, transcription defaults to en-US and misrecognises other languages. Note this property configures the source language only; profanity filtering is handled separately via ProfanityOption, so it is one of the three required settings.

Why this answer

Option C is correct because setting SpeechConfig.SpeechRecognitionLanguage specifies the recognition locale (for example, en-US or fr-FR) so the Speech SDK loads the appropriate acoustic and language model for the target language. Option D is correct because using the Speech SDK with intermediate results (via Recognizing events or continuous recognition) delivers real-time partial transcriptions as the caller speaks, which is required for live customer support transcription. Option E is correct because enabling the ProfanityFilter option (for example, setting Profanity to Masked or Removed in the SDK configuration) filters profane words from the recognized text.

Option A is not appropriate because the Batch Transcription REST API is asynchronous and designed for processing stored audio files, not real-time streaming. Option B is unnecessary because custom speech models are only needed to improve accuracy for domain-specific vocabulary or accents, not merely to handle multiple languages, which built-in models already support.

Exam trap

The trap is confusing batch and real-time APIs, and assuming custom models are needed for each language when built-in models suffice.

Full explanation →

544

MCQeasy

You need to translate a large batch of documents from English to multiple languages. Which Azure service should you use?

A.Azure AI Translator

B.Conversational Language Understanding

C.Azure AI Language Detection

D.Azure AI Speech Translation

AnswerA

Azure AI Translator provides a dedicated batch document translation capability that accepts whole documents in Blob Storage and renders them into multiple target languages, matching the large-batch, multi-language requirement. The synchronous single-string translate operation cannot process document batches at this scale.

Why this answer

Azure AI Translator is the correct service because it is specifically designed for batch document translation across multiple languages, supporting both text and document translation with source language auto-detection. It provides a dedicated Document Translation feature via the Translator API, which can handle large volumes of files asynchronously while preserving document structure and formatting.

Exam trap

The trap here is that candidates may confuse Azure AI Translator's batch document translation capability with Azure AI Speech Translation, mistakenly thinking speech translation can handle documents, or they may pick Language Detection because they assume detecting the source language is the primary need, overlooking the translation requirement.

How to eliminate wrong answers

Option B (Conversational Language Understanding) is wrong because it is designed for intent recognition and entity extraction from conversational utterances, not for translating document content between languages. Option C (Azure AI Language Detection) is wrong because it only identifies the language of a given text, without performing any translation. Option D (Azure AI Speech Translation) is wrong because it focuses on real-time translation of spoken audio streams, not on batch processing of written documents.

Full explanation →

545

MCQhard

Refer to the exhibit. A developer runs this PowerShell script to call Azure OpenAI. The script fails with an authentication error. What is the most likely cause?

A.The script uses the wrong HTTP header; it should use 'api-key' instead of 'Authorization: Bearer'.

B.The script uses the wrong HTTP method; it should use GET.

C.The API version is incorrect.

D.The endpoint URI is missing the resource name.

AnswerA

Azure OpenAI's data-plane REST API authenticates with the 'api-key' header carrying the resource key, not an OAuth bearer token. Sending 'Authorization: Bearer' therefore triggers a 401 authentication error, so switching the header to 'api-key' resolves the failure.

Why this answer

The script uses 'Authorization: Bearer' header, but Azure OpenAI requires the API key to be passed in the 'api-key' header. The 'Authorization: Bearer' header is used for Azure AD token-based authentication, not for direct API key authentication. Since the script is using an API key (as indicated by the PowerShell script), the correct header is 'api-key'.

Exam trap

The trap here is that candidates confuse Azure OpenAI's API key authentication with Azure AD token authentication, assuming 'Authorization: Bearer' is always correct, when in fact the header name differs based on the authentication method.

How to eliminate wrong answers

Option A is correct because Azure OpenAI API key authentication requires the 'api-key' header, not 'Authorization: Bearer'. Option B is wrong because the Azure OpenAI chat completions endpoint requires a POST method, not GET, to send the prompt and parameters in the request body. Option C is wrong because the API version is specified in the URI (e.g., '2023-12-01-preview') and an incorrect version would return a '400 Bad Request' or '404 Not Found', not an authentication error.

Option D is wrong because the endpoint URI includes the resource name (e.g., 'https://<resource>.openai.azure.com'), and a missing resource name would cause a DNS resolution failure or '404 Not Found', not an authentication error.

Full explanation →

546

Multi-Selectmedium

You are using Azure AI Language's question answering feature to build a bot that answers employee questions from a set of HR policy documents. You need to ensure the bot provides accurate answers and can handle follow-up questions. Which two actions should you take? (Choose two.)

Select 2 answers

A.Increase the confidence threshold to 90% to ensure only high-confidence answers are returned.

B.Enable multi-turn extraction to create follow-up prompts for related questions.

C.Use the exact match only setting to ensure answers are precise.

D.Add more source documents to the knowledge base to cover additional topics.

E.Add alternate questions to each question-answer pair to cover different phrasings.

AnswersB, E

Multi-turn extraction allows you to define follow-up prompts that guide the conversation, enabling the bot to ask clarifying questions or offer related answers. This is essential for handling follow-up questions and creating a more natural, interactive experience, especially for complex HR topics.

Why this answer

To improve accuracy and handle follow-ups in question answering, you should add alternate questions to capture different phrasings and enable multi-turn extraction to create follow-up prompts. These actions directly enhance the bot's ability to understand varied user input and maintain context. Other options either reduce flexibility or do not address the specific needs.

Exam trap

The trap here is assuming that raising the confidence threshold or adding more documents will improve accuracy and follow-up handling, when they can actually degrade the user experience.

Full explanation →

547

MCQmedium

You are reviewing an index definition created with PowerShell. The index is used for a knowledge mining solution that extracts people and organizations from documents. Users report that when they type partial names in the search bar, the suggester does not return suggestions. What is the most likely reason?

A.The people and organizations fields should be Edm.String instead of Collection(Edm.String)

B.The id field is not defined as a key in the index

C.The suggester sourceFields do not include people or organizations

D.The suggester searchMode should be 'analyzingInfixMatching' which is incorrect

AnswerC

The suggester only generates suggestions from the fields listed in its sourceFields collection. If people and organizations are absent from that list, partial-name queries return nothing, regardless of the analyser or searchMode settings. Adding those extracted fields to sourceFields satisfies the stem's requirement that partial names produce suggestions.

Why this answer

A suggester in Azure Cognitive Search only returns suggestions for fields explicitly listed in its `sourceFields` property. If the `people` and `organizations` fields are not included in `sourceFields`, the suggester cannot match partial names typed in the search bar, even if those fields are indexed and searchable. The suggester relies on prefix matching against the specified source fields to generate suggestions.

Exam trap

The trap here is that candidates may focus on data types or index keys instead of recognizing that the suggester's `sourceFields` property explicitly controls which fields participate in suggestion generation, a detail frequently tested in AI-102.

How to eliminate wrong answers

Option A is wrong because `Collection(Edm.String)` is the appropriate type for fields that contain multiple values (e.g., multiple people or organizations per document); changing them to `Edm.String` would lose multi-value support and is not related to suggester functionality. Option B is wrong because the `id` field being defined as a key is required for any index, but its presence or absence does not affect whether a suggester returns suggestions for other fields. Option D is wrong because `analyzingInfixMatching` is not a valid `searchMode` for a suggester; the correct `searchMode` is `analyzingInfixMatching` (note the typo in the option) but the real issue is that the suggester's `sourceFields` must include the target fields, not the search mode.

Full explanation →

548

MCQmedium

You are deploying an Azure AI solution that must process images stored in an Azure Blob Storage account. The solution uses the Computer Vision API and must be able to access the images without exposing storage account keys in code. You need to configure authentication. What should you do?

A.Store the storage account key in Azure Key Vault and retrieve it at runtime using the application's service principal.

B.Assign a managed identity to the Azure resource hosting the solution and grant it the Storage Blob Data Reader role on the storage account.

C.Use a shared access signature (SAS) token generated with the storage account key and embed it in the application configuration.

D.Enable anonymous read access on the blob container and configure the Computer Vision client to use the public URLs of the images.

AnswerB

Assigning a managed identity to the compute resource (e.g., Azure Function, VM) allows it to authenticate to Blob Storage without storing credentials. Granting the Storage Blob Data Reader role provides read access to blobs. This approach eliminates secrets in code and follows Azure security best practices for service-to-service authentication.

Why this answer

Using a managed identity with the appropriate RBAC role allows the solution to authenticate to Blob Storage without embedding secrets. The Storage Blob Data Reader role grants the necessary read access. This method is secure, requires no credential management, and aligns with Azure best practices for passwordless authentication.

Exam trap

The trap here is assuming that storing keys in Key Vault is as secure as using managed identities, but Key Vault still requires handling secrets at runtime.

Full explanation →

549

Multi-Selecteasy

Which TWO Azure AI services can you use to implement a custom question-answering system?

Select 2 answers

A.Azure OpenAI Service

B.Azure AI Bot Service

C.Azure AI Translator

D.Azure AI Language

E.Azure AI Search

AnswersA, D

Azure OpenAI Service supports custom question answering by grounding a model on your own data, typically via the On Your Data feature or retrieval augmentation, returning generated answers rather than only extracted passages. This satisfies the requirement for a custom question-answering system.

Why this answer

Azure OpenAI Service (A) is correct because it lets you build a custom question-answering system by grounding a chat/completions model on your own data (e.g., via the On Your Data feature or by supplying retrieved context), enabling natural-language answers over proprietary content. Azure AI Language (D) is correct because it includes the Custom Question Answering feature (formerly QnA Maker), which builds a knowledge base from documents, URLs, and question-answer pairs and exposes it through a REST API for question answering. Azure AI Bot Service (B) is a framework for hosting and connecting conversational bots, not a question-answering knowledge service, so it does not itself implement custom Q&A.

Azure AI Translator (C) only performs text translation between languages and has no question-answering capability. Azure AI Search (E) is a search/indexing service that can retrieve relevant documents but does not generate or manage question-answer pairs on its own, so it is not one of the two services for implementing custom question answering.

Exam trap

The trap here is that candidates often confuse Azure AI Search (a retrieval service) with a full question-answering system, forgetting that it only returns raw documents or passages and does not generate natural language answers, which requires a language model like Azure OpenAI Service or the custom question-answering feature in Azure AI Language.

Full explanation →

550

MCQhard

A manufacturing company uses Azure Custom Vision to detect defects on an assembly line. The model is deployed to a container on a local edge server. Recently, the model's accuracy dropped. You suspect data drift. What should you do to monitor and retrain the model?

A.Use Azure Machine Learning data drift monitoring on the Custom Vision endpoint.

B.Periodically collect new images with labels, retrain the model in Custom Vision, and redeploy the updated container.

C.Configure Custom Vision to send alerts when drift is detected.

D.Enable active learning in Custom Vision to automatically retrain the model.

AnswerB

Data drift is countered by capturing fresh labelled images from the line, retraining in Custom Vision so the model learns current defect patterns, then exporting and redeploying the updated container to the edge server. This restores accuracy while keeping inference local.

Why this answer

Custom Vision models deployed to containers on edge devices do not expose a REST endpoint that Azure Machine Learning's data drift monitoring can directly access. The only way to detect drift and retrain is to periodically collect new labeled images from the production line, retrain the model in Custom Vision, and redeploy the updated container to the edge server.

Exam trap

The trap here is that candidates assume Azure Machine Learning's data drift monitoring works with any deployed model, but it specifically requires an Azure-hosted endpoint, not a local container, and Custom Vision lacks native drift detection or auto-retraining features.

How to eliminate wrong answers

Option A is wrong because Azure Machine Learning data drift monitoring requires an Azure-hosted endpoint (e.g., AKS or ACI) with a scoring URI; Custom Vision containers on local edge servers do not provide such an endpoint, so drift monitoring cannot be configured. Option C is wrong because Custom Vision does not have built-in drift detection or alerting capabilities; it only provides training and prediction APIs, not monitoring. Option D is wrong because active learning in Custom Vision is a feature for image classification that suggests images for labeling to improve the model, but it does not automatically retrain the model or handle drift detection on edge deployments.

Full explanation →

551

MCQeasy

A healthcare organization uses Azure Document Intelligence to process patient intake forms. They notice that the confidence scores for field extraction are low. What is the most likely cause?

A.The document resolution is too low

B.The document layout is not analyzed

C.The custom model was trained with only 10 labeled forms

D.The batch processing size is too large

AnswerC

Custom models require at least 5 labeled forms; more samples improve confidence.

Why this answer

Custom models in Azure Document Intelligence require a minimum of five labeled forms for training, but low confidence scores typically indicate insufficient training data. With only 10 labeled forms, the model lacks enough examples to generalize well across variations in handwriting, formatting, and field values, leading to poor extraction confidence.

Exam trap

The trap here is that candidates often confuse low confidence with OCR or resolution issues, but the exam tests the specific requirement for sufficient labeled training data in custom models, not generic document quality problems.

How to eliminate wrong answers

Option A is wrong because low resolution can reduce OCR accuracy, but Azure Document Intelligence handles a wide range of resolutions and the question specifically points to field extraction confidence, not OCR failure. Option B is wrong because layout analysis is automatically performed by the prebuilt layout model and is not a prerequisite for custom extraction models; the issue is with training data quantity, not layout processing. Option D is wrong because batch processing size affects throughput and latency, not the confidence scores of individual field extractions; confidence is determined by the model's training and the input document quality, not batch size.

Full explanation →

552

MCQhard

Your team ships a generative assistant that calls a custom function to look up live order status. Users report that the assistant sometimes invents an order status instead of calling the function, and that when it does call the tool the arguments are occasionally malformed. You need to make tool invocation more reliable while keeping the existing model. What should you do?

A.Define the function with a clear name, description, and parameter schema, set tool_choice to require a tool call, and validate returned arguments before executing.

B.Switch the deployment to a larger model and keep the existing function definitions unchanged.

C.Add the phrase "always call the function" to the system message and leave tool_choice at its default value.

D.Raise the frequency_penalty so the model is discouraged from repeating previously invented order statuses.

AnswerA

Function calling reliability depends on precise tool metadata and an explicit tool_choice setting. Requiring a tool call stops the model from answering from memory, and a well-described parameter schema reduces malformed arguments. Validating arguments before execution prevents bad data from reaching the backend, which addresses both reported symptoms without changing the model.

Why this answer

Reliable function calling comes from three things working together: descriptive tool metadata so the model understands when and how to call it, an explicit tool_choice that forces a call when the intent is known, and client-side validation of the returned arguments before execution. Penalties, larger models, and prompt-only pleading do not enforce tool invocation or argument correctness.

Exam trap

The trap here is assuming that a stronger model or a sterner system message will force tool use, when the decision is governed by tool_choice and the argument shape by the parameter schema.

Full explanation →

553

MCQmedium

Refer to the exhibit. You are designing a Data Factory pipeline to perform sentiment analysis on a text column. The pipeline fails with a 'BadRequest' error. What is the most likely issue?

A.The output variable 'sentimentResult' is not defined

B.The activity type should be 'AzureFunction'

C.The input format is incorrect; it should be a JSON array of documents

D.The linked service name is misspelled

AnswerC

The Azure AI Language sentiment endpoint expects a request body shaped as a JSON array of document objects, each with an id, text and language. A BadRequest arises when the pipeline sends a flat string or malformed payload, so restructuring the input as that array satisfies the API's schema constraint.

Why this answer

The Cognitive Services activity in Azure Data Factory that performs sentiment analysis expects input in the form of a JSON array of documents, each containing an 'id' and 'text' field. A 'BadRequest' error typically indicates that the input format is incorrect, such as passing a single string or an improperly structured object instead of the required array. The other options are less likely because the error is specifically related to the request payload format.

Exam trap

Candidates often suspect configuration errors like linked service names or activity types, but the most common cause of BadRequest is an incorrect input format for the Cognitive Services activity.

How to eliminate wrong answers

Option A is wrong because the 'sentimentResult' output variable is defined in the activity's output mapping, and a missing definition would cause a different error (e.g., 'VariableNotFound') rather than a 'BadRequest' HTTP error. Option B is wrong because the activity type should be 'AzureMLBatchExecution' for invoking an Azure Machine Learning web service, not 'AzureFunction', which is used for Azure Functions. Option D is wrong because a misspelled linked service name would result in a 'LinkedServiceNotFound' or connection error, not a 'BadRequest' error from the service endpoint.

Full explanation →

554

MCQmedium

You are building an Azure AI Language solution that must extract named entities from support tickets and classify each entity as a person, organization, or location. The tickets are stored as UTF-8 text files. You need to call the REST API for Named Entity Recognition (NER) and ensure the response includes entity categories. Which request should you send?

A.POST to https://<resource>.cognitiveservices.azure.com/text/analytics/v3.1/entities/recognition/general with the documents array.

B.POST to https://<resource>.cognitiveservices.azure.com/language/:analyze-text?api-version=2023-04-01 with a JSON body containing "kind": "KeyPhraseExtraction" and the documents array.

C.GET to https://<resource>.cognitiveservices.azure.com/language/:analyze-text?api-version=2023-04-01 with the text as a query parameter.

D.POST to https://<resource>.cognitiveservices.azure.com/language/:analyze-text?api-version=2023-04-01 with a JSON body containing "kind": "EntityRecognition" and the documents array.

AnswerD

The Azure AI Language analyze-text endpoint accepts a POST request with the kind parameter set to EntityRecognition. The api-version 2023-04-01 is a valid version that returns entity categories such as Person, Organization, and Location. This matches the requirement to extract and categorize named entities from support tickets.

Why this answer

The correct request uses the unified Language service analyze-text endpoint with the EntityRecognition task, which returns entities with categories such as Person, Organization, and Location. The older Text Analytics endpoint is deprecated, and other task types like KeyPhraseExtraction do not provide entity categorization. Using POST with a JSON body is mandatory for analyze-text.

Exam trap

The trap here is assuming that any endpoint under the cognitive services domain will work, when only the unified analyze-text endpoint with the correct task kind returns categorized entities.

Full explanation →

555

MCQeasy

Refer to the exhibit. You are using Microsoft Graph to retrieve user information for use in a Microsoft 365 Copilot extension. The response shows that the mail and mobilePhone fields are null. What is the most likely reason?

A.The user has not configured those properties in their Microsoft Entra ID profile.

B.The API call required additional permissions.

C.The user is a guest user.

D.The user does not exist in the tenant.

AnswerA

Microsoft Graph returns only values populated in the directory; null mail and mobilePhone indicate those attributes were never set on the user object in Microsoft Entra ID. Missing consent or permissions would typically produce an error, not null fields.

Why this answer

The mail and mobilePhone fields are null because the user has not populated these attributes in their Microsoft Entra ID (formerly Azure AD) profile. Microsoft Graph returns the actual stored values for these properties; if they are empty or unset, the API response will show null. This is the most common and straightforward reason for null values in user profile fields.

Exam trap

Microsoft often tests the misconception that missing data in a successful API response is due to permission issues or user type, when in reality the most likely cause is that the data simply hasn't been configured.

How to eliminate wrong answers

Option B is wrong because if the API call required additional permissions, the response would return a 403 Forbidden error or an insufficient privileges message, not a successful response with null fields. Option C is wrong because guest users can have mail and mobilePhone properties configured; being a guest does not inherently cause these fields to be null—they would only be null if the guest user's profile lacks those values. Option D is wrong because if the user did not exist in the tenant, the API would return a 404 Not Found error, not a successful response with null fields.

Full explanation →

556

MCQmedium

You are implementing an Azure AI Search enrichment pipeline that extracts text from PDF documents stored in Azure Blob Storage. The PDFs are scanned images with no embedded text layer. You need to ensure the extracted text is available for downstream skills. Which skill should you add to the skillset?

A.Microsoft.Skills.Text.SplitSkill

B.Microsoft.Skills.Text.MergeSkill

C.Microsoft.Skills.Custom.WebApiSkill

D.Microsoft.Skills.Vision.OcrSkill

AnswerD

The OCR skill uses the Computer Vision Read API to extract text from images, including scanned PDF pages. It is designed for exactly this scenario where documents lack a text layer. The skill outputs text and layout information that can be mapped to index fields. This is the correct choice because it directly addresses the need to perform optical character recognition on image-based PDFs within the enrichment pipeline.

Why this answer

The OCR skill is specifically designed to extract text from images, including scanned PDFs, within an Azure AI Search enrichment pipeline. It leverages the Computer Vision Read API to recognize text and output it for further processing. Other skills like Split, Merge, or custom Web API do not provide built-in OCR capabilities.

Therefore, adding the OCR skill is the correct approach to make the scanned PDF content searchable.

Exam trap

The trap here is assuming that the built-in document extraction skill automatically handles scanned images, when in fact it only extracts text from documents with an embedded text layer.

Full explanation →

557

MCQeasy

Your chatbot uses Azure Bot Service and QnA Maker. Users can ask questions in natural language, and the bot returns answers from a knowledge base. Users report that the bot sometimes returns irrelevant answers. What should you do first?

A.Create multiple QnA Maker knowledge bases for different topics

B.Integrate LUIS to detect user intent

C.Use Azure AI Search to index the knowledge base

D.Review and edit the QnA pairs to add alternative phrasings

AnswerD

Editing QnA pairs to add alternative phrasings directly improves matching, because QnA Maker ranks answers by comparing the user's utterance against each question's stored wording. Irrelevant responses stem from weak lexical overlap, so enriching question variants raises confidence scores for the intended pair, satisfying the stem's requirement to fix irrelevant answers first.

Why this answer

The core issue is that the bot returns irrelevant answers because the QnA Maker knowledge base lacks sufficient alternative phrasings to match the variety of user questions. By reviewing and editing QnA pairs to add alternative phrasings, you directly improve the synonym and paraphrase coverage, which increases the confidence score for correct matches and reduces irrelevant responses. This is the first and most fundamental troubleshooting step before considering more complex integrations.

Exam trap

The trap here is that candidates often jump to integrating LUIS or Azure AI Search as a 'smart' fix, but the exam expects you to first optimize the existing QnA Maker knowledge base by enriching it with alternative phrasings, which is the simplest and most direct solution for irrelevant answers.

How to eliminate wrong answers

Option A is wrong because creating multiple knowledge bases for different topics does not address the root cause of irrelevant answers; it may fragment the knowledge and still fail to match varied phrasings within each topic. Option B is wrong because integrating LUIS for intent detection is an advanced enhancement that adds complexity and is not the first step; the problem is with QnA Maker's own matching logic, not with missing intent recognition. Option C is wrong because Azure AI Search is used for indexing and full-text search over large datasets, but QnA Maker already has its own ranking and matching engine; adding Azure AI Search would not fix the core issue of insufficient alternative phrasings in the QnA pairs.

Full explanation →

558

MCQhard

You are training an Azure Custom Vision object detection model to locate pallets in warehouse photos. Your training set contains 500 images, but only 40 images include pallets, while the rest are empty aisles. The model performs poorly, often missing pallets. You need to improve detection while keeping training time reasonable. What should you do first?

A.Enable the 'General' domain and retrain with the same dataset without changes.

B.Switch the project domain to a compact domain to speed up training.

C.Increase the probability threshold so fewer false positives are returned.

D.Add more labeled images that contain pallets in varied conditions and balance the dataset.

AnswerD

The model misses pallets because positive examples are scarce relative to empty aisles, so it learns the background class too strongly. Adding and labeling more pallet images across lighting, angles, and stacking arrangements gives the model representative positive features, improving recall. Balancing the dataset directly targets the root cause rather than masking it with threshold changes.

Why this answer

Object detection models learn from the distribution of labeled examples. When pallet images are only 8 percent of the set, the model is biased toward the dominant empty-aisle class and misses pallets. Adding more varied, labeled pallet images and balancing classes improves the positive signal.

Threshold tuning, domain changes, or simple retraining do not correct the underlying data imbalance.

Exam trap

The trap here is reaching for a threshold or domain tweak, when the real cause is too few labeled positive examples relative to empty background images.

Full explanation →

559

MCQmedium

You need to build a solution that reads text from images in multiple languages, including Arabic and English, and translates the text into English. The solution must preserve the original layout as much as possible. Which combination of Azure AI services should you use?

A.Azure AI Document Intelligence Read and Azure AI Translator

B.Azure AI Document Intelligence Read and Azure AI Language

C.Azure AI Vision OCR and Azure AI Translator

D.Azure AI Speech and Azure AI Translator

AnswerA

Azure AI Document Intelligence's Read model extracts printed and handwritten text with layout preserved as lines and words, supporting Arabic and English. Azure AI Translator then converts the extracted text into English. This combination satisfies both the multilingual OCR requirement and the layout-preservation constraint, which standalone Translator or Vision OCR cannot fully meet.

Why this answer

Azure AI Document Intelligence Read (formerly Form Recognizer Read) is optimized for extracting text from images and documents while preserving the original layout, including bounding box coordinates for each text element. Azure AI Translator then translates the extracted text into English. This combination meets the requirement for multi-language OCR (including Arabic and English) and layout preservation.

Exam trap

The trap here is that candidates often confuse Azure AI Vision OCR (legacy) with Azure AI Document Intelligence Read, assuming both provide equivalent layout preservation, but only Document Intelligence Read is designed for structured layout-aware extraction.

How to eliminate wrong answers

Option B is wrong because Azure AI Language provides text analytics (e.g., sentiment, key phrases) but does not include OCR capabilities; it cannot read text from images. Option C is wrong because Azure AI Vision OCR (legacy OCR API) does not preserve layout information as effectively as Document Intelligence Read, which is specifically designed for layout-aware extraction. Option D is wrong because Azure AI Speech is for speech-to-text and text-to-speech, not for reading text from images.

Full explanation →

560

MCQmedium

A company is building an agent using Azure AI Foundry Agent Service. The agent must be able to call an external REST API that returns real-time inventory data. The API requires an OAuth 2.0 token that changes frequently. Which approach should the team use to enable the agent to call this API securely?

A.Configure the agent to use a managed identity and assign it the necessary permissions to call the REST API directly.

B.Store the OAuth token in the agent's knowledge base and have the agent retrieve it when needed.

C.Use a function tool in the agent definition that invokes an Azure Function, which retrieves the token from Azure Key Vault and calls the REST API.

D.Embed the OAuth token in the agent's system prompt so the agent can include it in API calls.

AnswerC

This approach is correct because the agent can call an Azure Function as a function tool, and the function can securely retrieve the OAuth token from Key Vault at runtime. This keeps the token out of the agent's configuration and allows the function to handle token refresh and API calls, ensuring secure and up-to-date authentication for the external REST API.

Why this answer

The correct approach is to use a function tool that invokes an Azure Function, which securely retrieves the OAuth token from Azure Key Vault and calls the REST API. This keeps credentials out of the agent and handles token refresh. Other options either expose the token, rely on unsupported authentication, or misuse the knowledge base.

Exam trap

The trap here is assuming that managed identity can authenticate to any external API, but it only works with resources that trust Microsoft Entra ID.

Full explanation →

561

MCQhard

Your organization has a large corpus of legal documents that need to be analyzed for specific clauses. You need to extract key information such as party names, dates, and monetary amounts. The solution must be able to handle varying document formats (PDF, Word, scanned images). Which combination of Azure AI services should you use?

A.Azure AI Document Intelligence and Custom Entity Extraction

B.Azure AI Computer Vision and Custom Entity Extraction

C.Azure AI Translator and Custom Entity Extraction

D.Azure AI Speech and Custom Entity Extraction

AnswerA

Azure AI Document Intelligence handles PDF, Word and scanned images via its prebuilt and custom models, while Custom Entity Extraction trains on your labelled legal clauses to pull party names, dates and monetary amounts. Together they satisfy the varying-format constraint and the need for domain-specific field extraction.

Why this answer

Azure AI Document Intelligence (formerly Form Recognizer) is designed to extract text, structure, and key-value pairs from PDFs, Word documents, and scanned images using OCR and layout analysis. Combining it with Custom Entity Extraction (via Azure AI Language's custom NER) allows you to identify domain-specific entities like party names, dates, and monetary amounts across varying formats. This pairing directly addresses the need for both document parsing and tailored entity recognition.

Exam trap

The trap here is that candidates often confuse Azure AI Computer Vision's OCR capabilities with Document Intelligence's full document understanding, overlooking that Computer Vision lacks the ability to extract structured key-value pairs and custom entities without additional services.

How to eliminate wrong answers

Option B is wrong because Azure AI Computer Vision provides OCR and image analysis but lacks native support for extracting structured key-value pairs or custom entities from documents; it would require additional services to achieve the same result. Option C is wrong because Azure AI Translator is designed for language translation, not for extracting entities or analyzing document content, making it irrelevant for clause analysis. Option D is wrong because Azure AI Speech handles audio-to-text transcription and is not applicable to processing static document formats like PDFs, Word files, or scanned images.

Full explanation →

562

Multi-Selecteasy

Which TWO Azure AI services are most appropriate for extracting text from images and recognizing handwritten text?

Select 2 answers

A.Azure AI Document Intelligence

B.Azure AI Speech

C.Azure AI Vision

D.Azure AI Search

E.Azure AI Language

AnswersA, C

Azure AI Document Intelligence reads text from images and PDFs, and its Read model handles both printed and handwritten content. That dual capability satisfies the stem's requirement to extract text from images and recognise handwriting in one service.

Why this answer

Azure AI Document Intelligence (option A) is correct because it provides prebuilt and custom models that perform OCR and extract text, key-value pairs, and tables from documents, including support for handwritten text via its Read and Layout models. Azure AI Vision (option C) is correct because its Image Analysis and OCR capabilities, including the Read API, are specifically designed to extract printed and handwritten text from images. Azure AI Speech (option B) is incorrect because it handles speech-to-text and text-to-speech, not text extraction from images.

Azure AI Search (option D) is incorrect because it is a search indexing and query service, not an OCR or handwriting recognition service. Azure AI Language (option E) is incorrect because it focuses on natural language processing tasks such as sentiment analysis, entity recognition, and translation, not image text extraction.

Exam trap

AI-102 often tests whether candidates conflate text-analytics services (Language, Search) with OCR services — the exam wants the two services that actually perform image and handwriting text extraction.

Full explanation →

563

Multi-Selecthard

Your organization is using Azure AI Document Intelligence to process a mix of invoices and purchase orders. You need to ensure that documents are correctly classified before extraction. Which THREE steps should you take?

Select 3 answers

A.Train the classification model with one sample per type

B.Create a custom classification model in Document Intelligence

C.Label at least 5 samples for each document type

D.Chain the classification model with extraction models

E.Use the prebuilt invoice and purchase order models for classification

AnswersB, C, D

A custom classification model in Document Intelligence identifies each document's type before extraction, letting the pipeline route invoices and purchase orders to the appropriate extraction model. This directly satisfies the requirement to classify correctly beforehand.

Why this answer

Option B is correct because a custom classification model in Azure AI Document Intelligence is the required resource for identifying document type before extraction, since prebuilt models extract fields but do not perform custom document classification. Option C is correct because training a custom classification model requires a minimum of 5 labeled samples per document type, which provides the model with enough examples to distinguish invoices from purchase orders. Option D is correct because after classification, you must chain the classification model with the appropriate extraction models so each document is routed to the correct extraction model for field retrieval.

Option A is incorrect because one sample per type is insufficient; the minimum is 5 labeled samples per document type. Option E is incorrect because prebuilt invoice and purchase order models are extraction models, not classification models, and they do not classify documents before extraction.

Exam trap

The trap here is that candidates confuse prebuilt models (which perform extraction) with classification capabilities, assuming they can automatically identify document types without a dedicated classifier.

Full explanation →

564

MCQhard

You manage an Azure AI Search service that indexes legal documents. The search latency is high, and you need to improve query performance without reducing index size. Which action should you take?

A.Upgrade to a higher pricing tier

B.Increase the number of partitions

C.Reduce the number of searchable fields

D.Increase the number of replicas

AnswerD

Replicas serve query execution, so adding them distributes search load across more nodes and lowers latency while leaving index size untouched. Partitions would increase storage and index capacity instead, which the stem explicitly rules out.

Why this answer

Increasing the number of replicas distributes query load across multiple copies of the index, which directly improves query throughput and reduces latency. Replicas are designed for scaling query operations without changing the index size or storage capacity.

Exam trap

The trap here is that candidates often confuse partitions (which scale storage and indexing) with replicas (which scale query performance), leading them to incorrectly choose increasing partitions when the real need is to reduce query latency.

How to eliminate wrong answers

Option A is wrong because upgrading to a higher pricing tier increases both storage and compute capacity, but it is an overkill when the goal is specifically to improve query performance without reducing index size; partitions are the correct scaling unit for storage and indexing throughput. Option B is wrong because increasing the number of partitions improves indexing throughput and storage capacity, not query latency; partitions do not help with query concurrency or response time. Option C is wrong because reducing the number of searchable fields would shrink the index size, which violates the requirement to not reduce index size, and it may degrade search relevance rather than directly address query latency.

Full explanation →

565

MCQeasy

You need to deploy a generative AI model that can generate images from text descriptions. Which Azure service should you use?

A.Azure OpenAI Service

B.Azure Machine Learning

C.Azure AI Vision

D.Azure AI Language

AnswerA

Azure OpenAI Service hosts DALL-E models that generate images directly from text prompts, satisfying the text-to-image requirement. Unlike Azure AI Vision, which only analyses or classifies existing images, it performs true generative synthesis. This makes it the appropriate service for producing original visuals from descriptions.

Why this answer

Azure OpenAI Service provides access to advanced generative AI models like DALL-E, which are specifically designed to generate images from natural language text descriptions. This service offers pre-trained models that can create high-quality images based on textual prompts, making it the correct choice for this task.

Exam trap

The trap here is that candidates may confuse Azure AI Vision (which analyzes images) with image generation, or assume that Azure Machine Learning is the only way to implement generative AI, overlooking the purpose-built Azure OpenAI Service for this task.

How to eliminate wrong answers

Option B is wrong because Azure Machine Learning is a platform for building, training, and deploying custom machine learning models, not a pre-built service for generating images from text; it would require you to develop and train your own image generation model from scratch. Option C is wrong because Azure AI Vision is designed for analyzing and extracting information from images (e.g., object detection, OCR), not for generating images from text descriptions. Option D is wrong because Azure AI Language focuses on natural language processing tasks such as text analysis, translation, and sentiment analysis, and does not include image generation capabilities.

Full explanation →

566

MCQmedium

You are deploying a conversational AI chatbot using Azure AI Language service. The chatbot must be able to switch between multiple intents in a single conversation without restarting the session. Which feature should you enable?

A.Active learning

B.Orchestration workflow

C.Dynamic entity extraction

D.Prebuilt domain components

AnswerB

Orchestration workflow connects multiple Azure AI Language projects, such as conversational language understanding and question answering, into one bot. It routes each utterance to the appropriate intent mid-conversation, letting the chatbot switch intents without restarting the session.

Why this answer

Orchestration workflow in Azure AI Language service allows a chatbot to switch between multiple intents within a single conversation by routing requests to different language models or custom question answering knowledge bases. This enables the chatbot to handle diverse intents seamlessly without restarting the session, as each intent can be processed by the most appropriate component.

Exam trap

The trap is confusing active learning (a feedback loop for model improvement) with orchestration workflow (a routing mechanism for multi-intent conversations). Orchestration is the correct feature for switching intents without restarting.

How to eliminate wrong answers

Option B is wrong because orchestration workflow is used to connect multiple language models or services (e.g., combining LUIS with QnA Maker) but does not inherently enable switching between intents within a single conversation without restarting; it manages routing between different models. Option C is wrong because dynamic entity extraction handles the identification of entities that vary in value (e.g., dates, numbers) but does not affect the ability to switch between intents mid-conversation. Option D is wrong because prebuilt domain components provide ready-made models for common scenarios (e.g., booking flights) but do not enable dynamic intent switching; they are static and require retraining to adapt to new intents.

Full explanation →

567

Multi-Selecthard

Your organization is deploying a generative AI chatbot using Azure OpenAI Service. The chatbot must answer questions based on internal documents stored in Azure Blob Storage. You need to implement a retrieval-augmented generation (RAG) solution. Which THREE components are required? (Select THREE.)

Select 3 answers

A.Azure Functions for preprocessing

B.Azure AI Search index

C.Azure OpenAI On Your Data configuration

D.Azure SQL Database for metadata

E.Embedding model deployment in Azure OpenAI

AnswersB, C, E

Stores embeddings and enables vector search.

Why this answer

Azure AI Search is the core indexing and retrieval engine in a RAG solution. It ingests documents from Azure Blob Storage, creates a searchable index, and enables vector or hybrid search to retrieve relevant chunks. The chatbot then uses these retrieved chunks as context for the Azure OpenAI model to generate grounded answers.

Exam trap

The trap here is that candidates often confuse optional preprocessing components (like Azure Functions) or auxiliary storage (like Azure SQL Database) as mandatory, when the three essential pillars are the search index, the embedding model, and the Azure OpenAI On Your Data integration that ties retrieval to generation.

Full explanation →

568

MCQmedium

A company uses Azure Form Recognizer to extract data from invoices. The extracted data contains many errors for a specific vendor's invoices. What should they do?

A.Use a different prebuilt model.

B.Disable the OCR step.

C.Increase the confidence threshold.

D.Custom train a model with labeled examples of that vendor's invoices.

AnswerD

A custom model trained on labelled examples of that vendor's invoice layout teaches Form Recognizer the specific field positions and terminology it misreads, unlike the prebuilt invoice model. This directly addresses the vendor-specific extraction errors by adapting the model to that template's structure.

Why this answer

Azure Form Recognizer's prebuilt invoice model may not generalize well to vendor-specific layouts or data formats. By custom training a model with labeled examples of that vendor's invoices, you adapt the extraction to the unique fields, tables, and formatting, significantly reducing errors. This leverages the service's supervised learning capability to improve accuracy for domain-specific documents.

Exam trap

The trap here is that candidates assume increasing the confidence threshold (Option C) will fix extraction errors, but it only filters results rather than improving the underlying model's accuracy for vendor-specific formats.

How to eliminate wrong answers

Option A is wrong because using a different prebuilt model (e.g., from receipt to invoice) would not address vendor-specific variations; all prebuilt models are trained on generic datasets and lack customization for a single vendor's patterns. Option B is wrong because disabling the OCR step would prevent text extraction entirely, making data capture impossible; OCR is a foundational step in Form Recognizer's pipeline. Option C is wrong because increasing the confidence threshold only filters out low-confidence results, it does not correct extraction errors; it may reduce false positives but will not improve the model's ability to correctly parse vendor-specific fields.

Full explanation →

569

MCQhard

You are a machine learning engineer at a large retail company. The company has thousands of product descriptions that need to be updated regularly. They currently use a manual process. You propose using Azure OpenAI to generate new descriptions based on product attributes. You have a dataset of existing product descriptions and attributes stored in an Azure SQL Database. The solution must be cost-effective, scalable, and must not require retraining the model. You need to design the solution. You have the following options: Option A: Use Azure OpenAI with few-shot learning by including examples in the prompt for each product. Deploy the model on an Azure Kubernetes Service (AKS) cluster for high throughput. Option B: Use Azure OpenAI with prompt templates that include product attributes and call the API for each product. Use Azure Logic Apps to orchestrate the workflow and store results back to Azure SQL Database. Option C: Fine-tune a custom model on the existing product descriptions and deploy it as a managed endpoint. Use Azure Data Factory to batch process all products. Option D: Use Azure OpenAI with the batch API to generate descriptions for all products at once, using a single prompt that lists all products and attributes. Store the batch output in Azure Blob Storage and then import into Azure SQL Database. Which option should you choose?

A.Option C

B.Option D

C.Option A

D.Option B

AnswerD

Prompt templates with attributes are cost-effective and scalable.

Why this answer

(Azure Logic Apps) is the correct choice. It uses Azure OpenAI with prompt templates that insert product attributes, making individual API calls per product. This approach is scalable because Azure Logic Apps can handle high volumes with built-in retry and concurrency, and it is cost-effective as you only pay per API call and execution.

It does not require model retraining. In contrast, Option A (AKS) introduces unnecessary infrastructure complexity; Option C (fine-tuning) requires retraining; and Option D (batch API) risks exceeding prompt size limits and is less suitable for incremental updates.

Exam trap

The trap is that candidates may mistakenly choose Option A (AKS) thinking it provides better scalability, Option C (fine-tuning) for customization, or Option D (batch API) for efficiency, but they overlook that Option B (Azure Logic Apps) offers the right balance of cost-effectiveness, scalability, and no retraining for incremental updates.

How to eliminate wrong answers

Option A is wrong because few-shot learning with examples in the prompt for each product is not cost-effective for thousands of products—it increases token usage and latency, and deploying on AKS adds unnecessary infrastructure complexity without addressing the need for batch processing. Option B is wrong because using Azure Logic Apps to call the API for each product individually is not scalable for thousands of products—it would result in high latency, cost, and potential throttling, and it does not leverage batch processing for efficiency. Option C is wrong because fine-tuning a custom model requires retraining, which violates the requirement that the solution must not require retraining the model, and deploying as a managed endpoint adds ongoing cost and complexity.

Full explanation →

570

MCQhard

Your Azure AI Language custom question answering project answers product questions from a knowledge base. Users report that the bot returns the same generic answer regardless of how they phrase a question, even when the knowledge base contains the correct information. You need to diagnose why the model is not matching semantically similar user phrasings to the right answer. What should you do first?

A.Enable active learning and wait for the model to retrain automatically on user queries

B.Add alternate questions to each question-answer pair in the project

C.Review the project's default answer and confirm the deployed model is the one being queried by the bot

D.Increase the number of documents in the knowledge base and re-index the project

AnswerC

A generic answer returned for every query strongly suggests the runtime is falling back to the default answer, which happens when no question-answer pair meets the confidence threshold or when the bot queries a stale or empty deployment. Verifying the default answer text and confirming the bot targets the correct deployed model version is the fastest way to isolate a configuration or deployment mismatch before changing content.

Why this answer

When every phrasing produces the same generic response, the runtime is almost certainly falling back to the project's default answer, which occurs when no question-answer pair scores above the threshold or when the bot queries a deployment that lacks the trained model. Inspecting the default answer and verifying the deployment target isolates configuration problems before any content changes are made. Content remedies such as alternate questions or more documents address different symptoms.

Exam trap

The trap here is jumping to content fixes like alternate questions or more documents when the uniform generic answer is a strong signal of a default-answer fallback or a stale deployment target.

Full explanation →

571

MCQeasy

A company is building a knowledge mining solution using Azure AI Search. They need to extract text from handwritten notes stored as images in Azure Blob Storage. They want to use the built-in OCR skill in a skillset. Which cognitive service does the OCR skill rely on?

A.Azure AI Language

B.Azure AI Document Intelligence

C.Azure AI Translator

D.Azure AI Vision

AnswerD

This is correct because the OCR skill in Azure AI Search uses the Computer Vision OCR API from Azure AI Vision. It extracts text from images, including handwritten text, depending on the language and model. The skill sends image data to the Azure AI Vision service and receives recognized text, which is then enriched into the search index.

Why this answer

The built-in OCR skill in Azure AI Search is powered by the Computer Vision OCR API, which is part of Azure AI Vision. This service extracts text from images, including handwritten text, and returns it as structured data. The skill then makes this text available for indexing.

Other cognitive services like Language, Document Intelligence, and Translator serve different purposes and are not used by the OCR skill.

Exam trap

The trap here is confusing Azure AI Document Intelligence with the OCR skill, but Document Intelligence is a separate service not used by the built-in OCR skill.

Full explanation →

572

MCQeasy

You are building a solution to extract key phrases from customer reviews using Azure AI Language. Which feature should you use?

A.Sentiment Analysis

B.Language Detection

C.Key Phrase Extraction

D.Named Entity Recognition

AnswerC

Key Phrase Extraction returns the salient terms from supplied text, which is precisely the extraction task described. It is a prebuilt Azure AI Language feature, so no training data or custom model is needed to surface the main topics in each review.

Why this answer

Key Phrase Extraction is the correct feature because it is specifically designed to identify and return the main talking points or important terms from unstructured text, such as customer reviews. Azure AI Language's Key Phrase Extraction API analyzes the text structure and linguistic patterns to surface the most relevant phrases, which directly addresses the requirement to extract key phrases from reviews.

Exam trap

The trap here is that candidates often confuse Named Entity Recognition with Key Phrase Extraction, because both involve identifying important words, but NER is strictly for predefined entity types (e.g., person, location) while Key Phrase Extraction captures any salient topic or concept from the text.

How to eliminate wrong answers

Option A is wrong because Sentiment Analysis evaluates the emotional tone (positive, negative, neutral) of text, not the extraction of key phrases. Option B is wrong because Language Detection identifies the language in which the text is written (e.g., English, Spanish), not the key phrases within it. Option D is wrong because Named Entity Recognition identifies and categorizes entities like people, organizations, locations, and dates, but does not extract general key phrases or talking points from the text.

Full explanation →

573

MCQhard

You are a data scientist at a healthcare company. You have deployed a GPT-4 model using Azure OpenAI to answer patient inquiries about medical conditions. The model is configured with temperature=0.3 and max_tokens=200. Recently, the compliance team flagged that some responses contain contradictory information compared to the official medical guidelines. You need to ensure the model's answers align strictly with the provided medical documents (stored as PDFs in Azure Blob Storage). You have access to Azure Cognitive Search and Azure AI Document Intelligence. The solution must minimize hallucinations and not require retraining the model. What should you do?

A.Use prompt engineering to add a system message that tells the model to only answer based on the uploaded PDFs. Keep the current deployment.

B.Index the medical PDFs into Azure Cognitive Search. Configure the Azure OpenAI deployment to use 'Add your data' pointing to this index. Set the system message to instruct the model to base answers only on the retrieved context.

C.Fine-tune GPT-4 on the medical documents using Azure OpenAI fine-tuning capabilities. Use the fine-tuned model for the chatbot.

D.Deploy Azure AI Content Safety to filter responses that contradict guidelines. Set up a custom content filter using a list of approved phrases.

AnswerB

Indexing the PDFs in Azure Cognitive Search and wiring the deployment's "Add your data" to that index implements retrieval-augmented generation: relevant guideline passages are injected into the prompt, grounding each response. The system message restricting answers to retrieved context directly satisfies the no-retraining constraint while minimising contradictory output.

Why this answer

It uses Azure Cognitive Search to index the medical PDFs and then configures the Azure OpenAI deployment with 'Add your data' to retrieve relevant context from that index at inference time. This retrieval-augmented generation (RAG) approach grounds the model's answers in the official documents without retraining, directly addressing the compliance team's requirement to align responses with the provided guidelines and minimize hallucinations.

Exam trap

Microsoft often tests the distinction between prompt engineering (which is lightweight but unreliable for grounding) and RAG with a search index (which provides verifiable, document-grounded responses), leading candidates to choose the simpler prompt-only solution without considering its inability to enforce factual accuracy.

How to eliminate wrong answers

Option A is wrong because prompt engineering alone cannot guarantee that the model will only use the uploaded PDFs; the model's internal knowledge may still produce contradictory information, and there is no mechanism to enforce retrieval of the actual document content. Option C is wrong because fine-tuning GPT-4 on the medical documents would require retraining the model, which contradicts the requirement to not retrain, and fine-tuning does not inherently prevent hallucinations when the model encounters out-of-distribution queries. Option D is wrong because Azure AI Content Safety with a custom filter of approved phrases is a post-hoc filtering approach that cannot ensure the model's responses are grounded in the specific PDFs; it would only block or flag responses that match a predefined list, not align answers with dynamic document content.

Full explanation →

574

MCQeasy

You are building a mobile app that uses Azure AI Vision to generate captions for photos taken by users. The app must work offline in areas with no internet connectivity. Which Azure AI Vision feature should you use?

A.Azure AI Vision Docker container for Image Analysis

B.Azure AI Custom Vision

C.Azure AI Vision Read API

D.Azure AI Vision Image Analysis API

AnswerA

Azure AI Vision provides Docker containers that can run the Image Analysis capabilities locally, including caption generation. These containers can be deployed on-premises or on edge devices, enabling offline operation. By using the container, the mobile app can communicate with a local endpoint without internet, satisfying the offline requirement.

Why this answer

The Azure AI Vision Docker container for Image Analysis allows you to run the captioning feature locally, enabling offline operation. This is the only option that provides the required functionality without internet connectivity, making it the correct choice for a mobile app that must work in areas with no network.

Exam trap

The trap here is overlooking that Docker containers can run AI services locally, while the cloud APIs require connectivity.

Full explanation →

575

MCQmedium

You are troubleshooting an Azure AI Search indexer that fails to index a PDF file stored in Azure Blob Storage. The error message indicates that the document is encrypted. What is the most likely cause and solution?

A.The indexer is not configured with the PDF parser; set the parsing mode

B.The file format is unsupported; convert to PDF/A

C.The file is too large; split it into smaller parts

D.The PDF is encrypted; remove encryption before indexing

AnswerD

Azure AI Search's document cracking cannot decrypt password-protected or rights-managed PDFs, so the indexer reports the document as encrypted and skips it. Removing encryption before indexing, or supplying an unencrypted copy, lets the blob indexer extract text successfully.

Why this answer

The most likely cause is that the PDF is encrypted, and the solution is to remove encryption before indexing. Azure AI Search indexers cannot decrypt password-protected or encrypted PDFs; they require unencrypted content to extract text. Therefore, the file must be decrypted prior to indexing.

Exam trap

AI-102 often tests the assumption that indexer errors are due to configuration or format issues, but encrypted files require explicit decryption, which is a common oversight.

How to eliminate wrong answers

Option A is wrong because the error explicitly indicates encryption, not a missing PDF parser; the parser is likely configured correctly. Option B is wrong because PDF/A is a format for archiving, not a solution for encryption; the file format is supported. Option C is wrong because file size would produce a different error, not an encryption error.

Full explanation →

576

MCQmedium

You deploy a chat application using Azure OpenAI Service. Users report that the model sometimes generates inappropriate content. You need to implement a safety system that can be customized for your organization's policies. What should you use?

A.Use the content filter system in Azure OpenAI Studio

B.Use Azure AI Content Safety with custom categories and severity thresholds

C.Apply responsible AI templates from Azure AI Studio

D.Configure Microsoft Entra ID Conditional Access policies

AnswerB

Azure AI Content Safety provides configurable harm categories with adjustable severity thresholds, letting you align filtering with your organisation's own policies rather than relying on the model's fixed default filters. This directly satisfies the requirement for a customisable safety system.

Why this answer

Azure AI Content Safety provides a customizable content moderation service that allows you to define custom categories and severity thresholds aligned with your organization's specific policies. This enables you to filter inappropriate content beyond the default filters, giving you granular control over what the model generates.

Exam trap

The trap here is that candidates often confuse the built-in content filters in Azure OpenAI Studio (which are not customizable) with Azure AI Content Safety (which is a separate, customizable service), leading them to choose option A.

How to eliminate wrong answers

Option A is wrong because the content filter system in Azure OpenAI Studio provides only predefined content categories (e.g., hate, violence) with fixed severity levels, and cannot be customized to match an organization's unique policies. Option C is wrong because responsible AI templates in Azure AI Studio are design patterns and guidelines for building ethical AI, not a runtime content safety system that can filter generated outputs. Option D is wrong because Microsoft Entra ID Conditional Access policies control authentication and access to resources, not the content generated by the AI model.

Full explanation →

577

MCQmedium

Your organization is implementing a knowledge mining solution for a research institute that needs to extract chemical compound names and reactions from scientific articles in PDF format. The solution must use a custom model because the scientific terminology is not covered by built-in skills. You have trained a custom model using Azure AI Language's custom entity recognition (NER) and deployed it as a REST endpoint. You are using Azure AI Search with a skillset. How should you integrate the custom NER model into the enrichment pipeline?

A.Create a custom skill that calls the custom NER endpoint and map the output to the index fields.

B.Use a Language Understanding (LUIS) app to extract entities and call it from a custom skill.

C.Use the built-in Entity Recognition skill and configure it with your custom model's endpoint.

D.Configure the indexer to call the custom NER endpoint directly during indexing.

AnswerA

A custom skill in the skillset invokes the deployed custom NER REST endpoint, passing enriched document text and writing returned entities into index fields. Built-in skills cannot cover the scientific terminology, so wrapping the endpoint as a custom skill is the only integration path.

Why this answer

To integrate a custom NER model into an Azure AI Search enrichment pipeline, you must create a custom skill that calls the custom NER endpoint. The custom skill is a web API that the skillset invokes, and it can map the JSON output to index fields. This allows the enrichment pipeline to use the custom model's predictions.

Exam trap

The trap is assuming that built-in skills can be customized with a custom endpoint; candidates may choose the built-in Entity Recognition skill, but it does not support custom models.

How to eliminate wrong answers

Option B is wrong because LUIS is for intent recognition and conversational language understanding, not for custom entity recognition from text; it is not suitable for extracting chemical compound names. Option C is wrong because the built-in Entity Recognition skill uses a pre-trained model and cannot be configured with a custom model's endpoint; it does not support custom models. Option D is wrong because the indexer cannot directly call a custom NER endpoint; it must go through a skillset with a custom skill.

Full explanation →

578

MCQeasy

You need to analyze a video stream from a security camera to count the number of people entering a building. Which Azure AI service is most suitable?

A.Azure AI Spatial Analysis

B.Azure AI Custom Vision

C.Azure AI Computer Vision

D.Azure AI Video Indexer

AnswerA

Azure AI Spatial Analysis ingests live video and applies computer vision operations, including people counting and zone-based entry/exit tracking, directly on the stream. It satisfies the real-time counting constraint that image-only services cannot, since Azure AI Vision handles still images rather than continuous camera feeds.

Why this answer

Azure AI Spatial Analysis is the most suitable service because it is purpose-built for real-time spatial analysis of video streams, including counting people entering a building. It can process live camera feeds and detect when people cross a line or enter a zone, providing real-time counts. In contrast, Azure AI Video Indexer is designed for analyzing recorded video files, not live streams, making it less suitable for real-time security camera analysis.

Other options like Custom Vision and Computer Vision are for image classification and image analysis, respectively, not optimized for video streams.

Exam trap

The key trap is that candidates often choose Azure AI Video Indexer (Option D) because it is associated with video analysis, but it is designed for indexing and analyzing video files, not real-time streaming. Azure AI Spatial Analysis (Option A) is the correct choice for live video streams and people counting scenarios.

How to eliminate wrong answers

Option A is wrong because Azure AI Spatial Analysis is a feature within Azure AI Computer Vision that focuses on real-time spatial relationships and occupancy analysis (e.g., social distancing, people counting in a space) but is not optimized for video stream analysis with event-based counting like entering a building; it is more suited for static spatial monitoring. Option B is wrong because Azure AI Custom Vision requires training a custom model with labeled images to detect specific objects, which is overkill and less efficient for a standard people-counting task that can be handled by pre-built video analytics. Option C is wrong because Azure AI Computer Vision provides image analysis APIs (e.g., OCR, object detection) but lacks native video stream processing capabilities and event-based counting for people entering a building; it is designed for single-image analysis, not continuous video streams.

Full explanation →

579

MCQmedium

A travel agency wants its web app to summarize long customer complaint emails into a short paragraph that preserves the main points. The emails are in English and average 800 words. You need to use the extractive summarization feature of Azure AI Language and control how long the summary is. Which request parameter should you configure?

A.summaryLength

B.maxSentenceCount

C.sentenceCountToExclude

D.temperature

AnswerB

Extractive summarization selects the most representative sentences from the source text, and maxSentenceCount caps how many sentences are returned. Setting it produces a shorter or longer summary while keeping the original wording, which matches the requirement to control summary length for complaint emails while preserving the main points.

Why this answer

Extractive summarization returns the highest-ranked sentences from the input, so length is controlled by limiting how many sentences are returned. maxSentenceCount sets that cap and directly satisfies the requirement to produce a shorter or longer summary while retaining the original phrasing of the complaint email. The other parameters either belong to different tasks or do not exist on this request.

Exam trap

The trap here is assuming a generative-style length control like summaryLength exists for extractive summarization, when the task actually caps the number of extracted sentences.

Full explanation →

580

MCQhard

You are designing a generative AI solution that uses Azure OpenAI GPT-4 to answer customer support questions. The solution must comply with Microsoft's Responsible AI principles, particularly transparency and accountability. Which implementation approach best meets these requirements?

A.Use the model without any modifications, and have a human review all responses.

B.Fine-tune the model on a curated dataset of support tickets and disable content filtering.

C.Enable content filtering, log all interactions, and include a disclaimer that responses are AI-generated.

D.Use the default model deployment and rely on the model's inherent safety.

AnswerC

Logging every interaction creates the audit trail that accountability demands, while the AI-generated disclaimer delivers the transparency requirement by telling users they are not reading human output. Content filtering addresses harm prevention rather than the two named principles, but the logging and disclosure pairing directly satisfies the stem's stated constraints.

Why this answer

It directly addresses Microsoft's Responsible AI principles of transparency and accountability. Enabling content filtering (via Azure AI Content Safety) ensures harmful outputs are blocked, logging all interactions provides an audit trail for accountability, and including a disclaimer that responses are AI-generated satisfies transparency by clearly informing users they are interacting with an AI system.

Exam trap

The trap here is that candidates assume human review (Option A) or model fine-tuning (Option B) alone satisfy Responsible AI principles, but Microsoft explicitly requires automated content filtering, logging, and transparency disclaimers as part of a comprehensive compliance strategy.

How to eliminate wrong answers

Option A is wrong because using the model without modifications fails to implement content filtering or logging, leaving the solution non-compliant with accountability and safety requirements; human review alone is insufficient for real-time compliance and does not provide automated transparency. Option B is wrong because disabling content filtering violates safety principles, and fine-tuning on a curated dataset does not guarantee compliance with transparency or accountability; it also risks overfitting or introducing bias without proper oversight. Option D is wrong because relying solely on the model's inherent safety is insufficient; Azure OpenAI's default deployment does not enforce logging or disclaimers, and the model can still produce harmful or non-transparent outputs without explicit content filtering and audit mechanisms.

Full explanation →

581

Multi-Selecthard

You are building a knowledge mining solution with Azure AI Search. The solution indexes scanned PDF reports stored in Azure Blob Storage. Each report contains multiple embedded images with text that must be searchable. You have already created a data source, index, and indexer. You need to ensure that text from the embedded images is extracted and mapped to the 'content' field in the index. Which two actions should you perform? (Choose two.)

Select 2 answers

A.Enable image extraction by setting the 'imageAction' parameter to 'generateNormalizedImages' in the indexer definition.

B.Configure the indexer to use the 'text' parsing mode instead of the default 'json' mode.

C.Add an OCR skill to the skillset and set its context to /document/normalized_images/*.

D.Add an Entity Recognition skill to detect and extract text from images.

E.Add a Shaper skill that merges all image text into a single string before indexing.

AnswersA, C

This is correct because imageAction must be set to generateNormalizedImages to create normalized images from embedded images in the document. Without this, the OCR skill has no image inputs to process. The normalized images become available at /document/normalized_images/*, which is the required input path for the OCR skill.

Why this answer

To make text from embedded images searchable, you must first extract the images from the documents by setting imageAction to generateNormalizedImages. Then, you add an OCR skill with context /document/normalized_images/* to process each image and output the recognized text. This text can then be mapped to the content field.

Without both actions, the OCR skill either has no input or does not run per image.

Exam trap

The trap here is assuming that adding an OCR skill alone is sufficient, but without image extraction the OCR skill has no normalized images to process.

Full explanation →

582

MCQhard

An organization is deploying a conversational AI solution using Azure OpenAI. They want to ensure the model's responses are grounded in their own knowledge base documents to reduce hallucinations. Which approach should they implement?

A.Integrate Azure Cognitive Search for retrieval-augmented generation (RAG)

B.Fine-tune the model on the knowledge base documents

C.Implement Azure AI Content Safety filters

D.Use prompt engineering to instruct the model to only use the knowledge base

AnswerA

Retrieval-augmented generation queries Azure Cognitive Search to fetch relevant passages from the organisation's own documents, then passes them into the Azure OpenAI prompt as grounding context. This directly satisfies the requirement to reduce hallucinations by anchoring responses in the knowledge base rather than relying solely on parametric model memory.

Why this answer

Retrieval-Augmented Generation (RAG) with Azure Cognitive Search allows the model to dynamically retrieve relevant chunks from the organization's knowledge base documents at inference time. This grounds responses in authoritative, up-to-date content, directly reducing hallucinations by providing factual context rather than relying solely on the model's parametric memory.

Exam trap

The trap here is that candidates often confuse fine-tuning (B) as a way to 'teach' the model the knowledge base, not realizing that RAG is the recommended pattern for grounding responses in external, query-specific data without retraining.

How to eliminate wrong answers

Option B is wrong because fine-tuning adjusts the model's weights on a static dataset, which can lead to overfitting and does not guarantee that the model will reference the knowledge base for every query; it also fails to incorporate new or updated documents without retraining. Option C is wrong because Azure AI Content Safety filters only block harmful or inappropriate content after generation; they do not provide factual grounding or reduce hallucinations. Option D is wrong because prompt engineering alone cannot enforce factual adherence; the model may still generate plausible-sounding but incorrect information from its training data, as it lacks a retrieval mechanism to verify claims against the knowledge base.

Full explanation →

583

MCQmedium

You deploy an Azure AI Services resource using the ARM template shown in the exhibit. You need to test the Language service API from your local machine. What should you do first?

A.Configure a managed identity for the resource

B.Add your public IP address to the ipRules array in the networkAcls

C.Change the defaultAction to Allow

D.Use Azure CLI to enable the resource

AnswerB

The networkAcls ipRules array is the ARM property controlling which public addresses may reach the endpoint when default action is Deny. Adding your public IP permits local API calls without redeploying or disabling the firewall, satisfying the test-from-local-machine constraint.

Why this answer

The ARM template in the exhibit sets `defaultAction` to `Deny`, which blocks all traffic not explicitly allowed by the `ipRules` array. To test the Language service API from your local machine, you must add your public IP address to the `ipRules` array so that the resource's network firewall permits inbound requests from your IP. Without this step, all API calls from your local machine will be rejected with a 403 Forbidden error.

Exam trap

The trap here is that candidates assume changing `defaultAction` to `Allow` is the simplest fix, but the question tests understanding that the resource is already deployed and the firewall is blocking traffic—so the correct first step is to explicitly permit your specific IP, not to open the resource to the entire internet.

How to eliminate wrong answers

Option A is wrong because configuring a managed identity is used for authenticating Azure resources to each other (e.g., allowing a VM to access the Language service without keys), but it does not bypass the network firewall; the IP-based access control must still allow the request. Option C is wrong because changing `defaultAction` to `Allow` would open the resource to all internet traffic, which is a security risk and not the minimal required step; the question asks what you should do first, and adding your specific IP is the correct, least-privilege approach. Option D is wrong because using Azure CLI to enable the resource is unnecessary—the resource is already deployed and enabled via the ARM template; the issue is network access control, not resource provisioning.

Full explanation →

584

Multi-Selecthard

You are deploying a generative AI model using Azure AI Foundry. The model must be accessible only from a specific virtual network. Additionally, you need to monitor all API calls for auditing. Which two configurations are required? (Choose two.)

Select 2 answers

A.Enable diagnostic settings to send logs to a Log Analytics workspace.

B.Assign a managed identity to the model deployment.

C.Enable public network access from selected IP addresses.

D.Disable public network access and configure a private endpoint.

E.Configure CORS to allow only the VNet's domain.

AnswersA, D

Logs enable auditing of all API calls.

Why this answer

Enabling diagnostic settings to send logs to a Log Analytics workspace captures all API call details (e.g., request URI, response status, caller IP) for auditing and monitoring. This is the standard Azure method for collecting resource-level logs, and it works with Azure AI Foundry deployments to meet compliance and security requirements.

Exam trap

The trap here is that candidates confuse network access controls (private endpoints) with authentication mechanisms (managed identities) or browser-level restrictions (CORS), leading them to select B or E instead of the correct pairing of D and A.

Full explanation →

585

Multi-Selectmedium

A team is building an agent with Azure AI Foundry Agent Service that must call several internal function tools. The team reports that the model sometimes invents function names that do not exist and passes arguments that do not match the tool schema. Which TWO practices should the team adopt to reduce these failures? (Choose two.)

Select 2 answers

A.Disable parallel tool calls so the model can only invoke one function per turn

B.Set the agent's temperature to a high value so the model explores more tool combinations

C.Handle tool-call validation errors by returning a structured error message to the model and allowing it to retry

D.Define each tool with a clear name, description, and JSON schema for its parameters

E.Increase the model's max tokens so it has more room to explain each function call

AnswersC, D

When a tool call fails validation, returning a clear, structured error as the tool result lets the model correct its arguments on the next turn. This feedback loop turns a hard failure into a recoverable step and reduces the chance that the run aborts. It complements precise tool schemas by catching the residue of malformed calls that still slip through.

Why this answer

Reliable function calling depends on a precise contract and a recovery path. Clear names, descriptions, and JSON schemas tell the model exactly which tools exist and how to call them, while returning structured validation errors lets the model self-correct on a subsequent turn. Temperature, parallel-call settings, and token limits affect other behaviors and do not resolve hallucinated function names or malformed arguments.

Exam trap

The trap here is assuming model sampling settings control tool accuracy, when tool schema quality and validation feedback drive correct function invocation.

Full explanation →

586

MCQeasy

You need to detect if a photo contains adult or racy content. Which Azure AI Computer Vision feature should you use?

A.Describe Image API

B.OCR API

C.Analyze Image API with the 'adult' parameter

D.Tag Image API

AnswerC

The Analyze Image API accepts an adult parameter that returns adult and racy classification scores with confidence values. This directly satisfies the requirement to detect adult or racy content in a photo, unlike OCR or tagging features.

Why this answer

The Analyze Image API with the 'adult' parameter is the correct feature because it specifically detects adult, racy, and gory content in images. When you call the Analyze Image API and include the 'adult' visual feature, Azure AI Computer Vision returns a boolean flag and a confidence score for adult and racy content classification, enabling content moderation.

Exam trap

The trap here is that candidates often confuse the Tag Image API's generic object tagging with the specialized adult content detection feature, assuming tags like 'swimsuit' or 'underwear' would suffice, but only the Analyze Image API with the 'adult' parameter provides the explicit moderation scores required by the question.

How to eliminate wrong answers

Option A is wrong because the Describe Image API generates human-readable captions summarizing the image content, but it does not provide explicit adult/racy content detection or confidence scores. Option B is wrong because the OCR API extracts printed or handwritten text from images, and has no capability to analyze visual content for adult or racy themes. Option D is wrong because the Tag Image API returns a list of content tags (e.g., 'person', 'tree') based on objects and actions, but it does not include a dedicated adult/racy content moderation feature.

Full explanation →

587

MCQhard

You are deploying a Custom Vision model to a production environment. The model must handle 100 predictions per second with low latency. Which deployment option should you choose?

A.Use the Free tier prediction endpoint.

B.Export the model as a Docker container and run it on Azure Container Instances.

C.Use the Training API to make predictions.

D.Use a paid tier prediction endpoint with sufficient capacity.

AnswerD

A paid tier prediction endpoint supplies dedicated throughput and lower latency than the free tier, whose strict rate limits cannot sustain 100 predictions per second. Scaling the endpoint's capacity directly satisfies the stem's high-volume, low-latency constraint, making it the appropriate deployment choice for production Custom Vision inference.

Why this answer

A paid tier prediction endpoint in Azure Custom Vision is designed to handle production-scale workloads with dedicated compute resources, supporting up to 100 predictions per second with low latency. The Free tier is rate-limited and cannot sustain this throughput, while exporting as a Docker container introduces additional overhead and scaling complexity that may not guarantee the required latency or throughput without manual orchestration.

Exam trap

The trap here is that candidates may assume exporting a model as a Docker container (Option B) is always the best for performance, but they overlook the operational overhead and lack of built-in scaling for high-throughput cloud predictions, whereas the paid endpoint is optimized for exactly this scenario.

How to eliminate wrong answers

Option A is wrong because the Free tier prediction endpoint is rate-limited to 20 predictions per minute and cannot handle 100 predictions per second. Option B is wrong because exporting the model as a Docker container and running it on Azure Container Instances requires manual scaling and does not provide built-in load balancing or guaranteed low latency for high-throughput production workloads; it is better suited for offline or edge scenarios. Option C is wrong because the Training API is used for training and managing models, not for making real-time predictions; using it for predictions would be inefficient and unsupported.

Full explanation →

588

MCQeasy

A marketing team uses Azure OpenAI Service to draft product descriptions. They report that outputs vary widely in tone and sometimes ignore the required brand voice. You need to make responses follow a fixed set of style rules consistently across all requests without changing the model deployment. What should you do?

A.Create a second deployment of the same model with a different name

B.Raise the temperature value to increase output creativity

C.Add a system message that defines the brand voice and formatting rules

D.Set the presence_penalty parameter to a large positive value

AnswerC

The system message is a high-priority instruction that shapes the model's behavior across the conversation, making it the correct place to encode tone, persona, and formatting constraints. Applied to every request, it enforces the brand voice consistently without retraining or redeploying the model. This is the standard, low-cost way to steer generative output in Azure OpenAI Service.

Why this answer

System messages carry instructions the model treats with high priority and apply to the whole request, so they are the right mechanism for enforcing brand voice, tone, and formatting rules. Sampling parameters such as temperature and presence_penalty alter randomness and repetition, and additional deployments of the same model behave identically, so none of them can encode style policy.

Exam trap

The trap here is reaching for a sampling parameter to control style, when style guidance belongs in the system message.

Full explanation →

589

MCQmedium

Your company uses Microsoft 365 Copilot to generate meeting summaries. Some users report that summaries include information from meetings they did not attend. What is the most likely cause?

A.The meeting organizer granted everyone in the organization view access.

B.Users have access to meeting artifacts via shared calendars or transcripts.

C.Copilot is using Bing search results to augment summaries.

D.Copilot is incorrectly configured to ignore meeting permissions.

AnswerB

Copilot surfaces content from meeting artefacts the user can already reach through Microsoft 365 permissions, such as shared calendars, transcripts and recordings. Summaries therefore draw on meetings the user did not attend but still has access to, which is the most likely cause of the reported behaviour.

Why this answer

Microsoft 365 Copilot generates meeting summaries by aggregating content from meeting artifacts such as transcripts, recordings, and shared calendars. If a user has access to a meeting's transcript or recording (e.g., via a shared calendar or because the meeting was recorded and stored in a location the user can access), Copilot can include that meeting's information in summaries even if the user did not attend. This behavior is by design, as Copilot respects existing permissions on the underlying data.

Exam trap

The trap here is that candidates often assume Copilot uses meeting attendance or organizer permissions to filter summaries, when in fact it relies on the underlying permissions of the meeting's artifacts (transcripts, recordings, calendar items), which can be broader than the attendee list.

How to eliminate wrong answers

Option A is wrong because granting everyone view access to a meeting would allow users to see the meeting details, but Copilot does not automatically include meetings in summaries based solely on view access; it requires access to the meeting's artifacts like transcripts or recordings. Option C is wrong because Copilot does not use Bing search results to augment meeting summaries; it relies on the user's Microsoft Graph data and permissions, not external web searches. Option D is wrong because there is no configuration setting in Copilot to 'ignore meeting permissions'; Copilot strictly adheres to the permissions set on the meeting artifacts and does not have a mode that bypasses them.

Full explanation →

590

MCQmedium

Your organization is building a chatbot using Azure OpenAI Service. The chatbot must provide citations from a set of internal documents stored in Azure Blob Storage. You need to configure the solution to minimize token usage while ensuring citations are accurate. Which approach should you use?

A.Embed all document content into the system prompt

B.Fine-tune a model on the documents so it can recall them from memory

C.Use a large context window model (e.g., 32K) and include all documents in the prompt

D.Use Azure OpenAI on your data with Azure Cognitive Search for hybrid retrieval

AnswerD

Azure OpenAI on your data with Cognitive Search hybrid retrieval combines keyword and vector search, returning only the most relevant document chunks as grounding context. That narrows the prompt payload, minimising tokens while preserving accurate citations from the Blob Storage documents.

Why this answer

Azure OpenAI on your data with Azure Cognitive Search for hybrid retrieval combines vector search and keyword search to efficiently find relevant document chunks from Azure Blob Storage, minimizing token usage by only sending the most pertinent content to the model for citation generation. This approach ensures accurate citations without embedding all documents into the prompt or relying on model memory.

Exam trap

The trap here is that candidates often confuse fine-tuning with retrieval-augmented generation (RAG), assuming fine-tuning can store factual knowledge for citation, when in reality RAG with a search index is required for accurate, token-efficient document grounding.

How to eliminate wrong answers

Option A is wrong because embedding all document content into the system prompt would consume an enormous number of tokens, exceeding context limits and incurring high costs, while also being impractical for large document sets. Option B is wrong because fine-tuning a model on documents does not enable it to recall specific citations accurately; fine-tuning adjusts model behavior but does not store document content for retrieval, leading to hallucinations or incorrect references. Option C is wrong because using a large context window model (e.g., 32K) and including all documents in the prompt still wastes tokens on irrelevant content, increases latency and cost, and does not guarantee accurate citations as the model may lose focus on the specific source material.

Full explanation →

591

MCQmedium

You are using Azure OpenAI to generate product descriptions. You notice that the descriptions are often too similar to each other. Which parameter should you adjust to increase diversity?

A.Increase the temperature value.

B.Decrease the top_p value.

C.Increase the max_tokens value.

D.Increase the frequency_penalty value.

AnswerA

Raising temperature flattens the model's probability distribution over next tokens, so lower-probability words are sampled more often, producing varied outputs. This directly satisfies the stem's requirement to increase diversity in the generated product descriptions, countering the repetitive similarity observed.

Why this answer

Increasing the temperature parameter makes the model more creative by raising the probability of sampling lower-probability tokens, which increases diversity in the generated text. A higher temperature (e.g., 0.9) flattens the probability distribution, so the model is less likely to always pick the most probable next word, resulting in more varied outputs.

Exam trap

Microsoft often tests the distinction between temperature (which controls randomness/creativity) and frequency_penalty (which controls repetition), leading candidates to mistakenly choose frequency_penalty when the question asks for diversity in content rather than just avoiding repetition.

How to eliminate wrong answers

Option B is wrong because decreasing top_p (nucleus sampling) reduces the cumulative probability mass considered for token selection, which actually makes outputs less diverse by focusing only on the most likely tokens. Option C is wrong because increasing max_tokens only extends the maximum length of the generated response; it does not affect the randomness or diversity of token choices. Option D is wrong because increasing frequency_penalty reduces the likelihood of repeating the same tokens or phrases, which can increase lexical diversity but does not directly control the overall creativity or randomness of the output like temperature does.

Full explanation →

592

Multi-Selecthard

You are building a document processing solution that extracts information from invoices. The invoices come in various formats and languages. You need to extract line items, totals, and supplier names. Which THREE services should you combine?

Select 3 answers

A.Azure AI Custom Vision

B.Azure AI Translator

C.Azure AI Content Safety

D.Azure AI Document Intelligence

E.Azure AI Vision OCR

AnswersB, D, E

Translates text if invoices are in multiple languages.

Why this answer

Azure AI Translator is correct because invoices arrive in various languages, and translating extracted text to a common language (e.g., English) is necessary for downstream processing like entity extraction and validation. Without translation, multilingual invoice data would be inconsistent or unprocessable by language-specific models.

Exam trap

The trap here is that candidates may mistakenly choose Azure AI Custom Vision for 'extracting' invoice data, confusing its image classification capabilities with the structured document extraction provided by Document Intelligence.

Full explanation →

593

MCQeasy

Refer to the exhibit. An Azure Cognitive Services Computer Vision API call for image captioning is returning only one caption. The developer wants to get three possible captions ranked by confidence. Which parameter should be modified in the request?

A.Use a different API version, such as 2023-04-01.

B.Modify the URL to point to a different image.

C.Change the language parameter to 'multi'.

D.Set the maxCandidates value to 3.

AnswerD

The maxCandidates parameter controls how many alternative captions the Image Captioning service returns, each with a confidence score. Setting it to 3 satisfies the requirement for three ranked captions; leaving it at the default of 1 explains why only one caption currently appears.

Why this answer

The `maxCandidates` parameter in the Computer Vision Image Analysis API controls the maximum number of captions returned in the response. By default, this value is 1, so only the top-ranked caption is returned. Setting `maxCandidates=3` instructs the API to return up to three captions, each with its own confidence score, ranked from highest to lowest confidence.

Exam trap

The trap here is that candidates may confuse the `maxCandidates` parameter with other parameters like `language` or `details`, or assume that changing the API version or image source would increase the number of captions, when in fact the default behavior is to return only one caption unless explicitly overridden.

How to eliminate wrong answers

Option A is wrong because changing the API version (e.g., to 2023-04-01) does not affect the number of captions returned; the `maxCandidates` parameter is available across supported versions. Option B is wrong because pointing to a different image changes the input but does not alter the request parameter that controls the number of captions; the API would still return only one caption per image unless `maxCandidates` is set. Option C is wrong because the `language` parameter specifies the language of the returned text (e.g., 'en' for English), not the count of captions; 'multi' is not a valid language value for this API.

Full explanation →

594

MCQmedium

A company builds a knowledge mining solution using Azure AI Search with a custom skillset that includes an OCR skill. They want to ensure that images embedded in PDFs are processed. What should they configure?

A.Set the 'defaultLanguageCode' to 'en'

B.Set the 'textExtractionAlgorithm' to 'printed'

C.Set the 'imageAction' parameter to 'generateNormalizedImages'

D.Set the 'lineEnding' parameter to 'space'

AnswerC

This parameter enables extraction of images from documents.

Why this answer

The 'imageAction' parameter in Azure AI Search's OCR skill controls whether images embedded in documents (including PDFs) are extracted and processed. Setting it to 'generateNormalizedImages' ensures that images within PDFs are normalized and passed to the OCR skill for text extraction, which is essential for processing embedded images.

Exam trap

The trap here is that candidates may confuse parameters that affect OCR output formatting (like 'lineEnding' or 'defaultLanguageCode') with the parameter that actually enables image extraction from PDFs, leading them to overlook the 'imageAction' setting.

How to eliminate wrong answers

Option A is wrong because 'defaultLanguageCode' specifies the language for text recognition, not whether images are extracted from PDFs; it does not enable image processing. Option B is wrong because 'textExtractionAlgorithm' determines the OCR algorithm (e.g., 'printed' or 'handwritten') but does not control the extraction of images from PDFs; it only affects how text is recognized once images are available. Option D is wrong because 'lineEnding' parameter controls the line break character in OCR output (e.g., 'space', 'carriageReturn'), which is irrelevant to enabling image extraction from PDFs.

Full explanation →

595

MCQmedium

Refer to the exhibit. An administrator runs this Azure CLI command to deploy a GPT-4 model in Azure AI Foundry. The command fails with an error that the deployment name already exists. What should the administrator do to resolve the issue?

A.Use a different deployment name or delete the existing deployment.

B.Specify a different resource group.

C.Remove the --sku-name parameter.

D.Use a different model version.

AnswerA

Deployment names must be unique within an Azure AI Foundry resource. Since the CLI command failed because that name is already taken, the administrator must either supply a new unique name or remove the existing deployment before retrying.

Why this answer

The error message indicates that a deployment with the same name already exists in the Azure AI Foundry workspace. In Azure AI Foundry, deployment names must be unique within a workspace. The correct resolution is to either choose a different deployment name or delete the existing deployment before re-running the command.

This aligns with the Azure CLI behavior where resource names (including AI model deployments) must be unique per scope.

Exam trap

The trap here is that candidates may think the error is about model availability or SKU constraints, but the error explicitly states 'deployment name already exists,' which is a naming conflict, not a capacity or version issue.

How to eliminate wrong answers

Option B is wrong because specifying a different resource group does not resolve a deployment name conflict within the same workspace; the deployment name uniqueness is scoped to the workspace, not the resource group. Option C is wrong because removing the --sku-name parameter would change the pricing tier or capacity, but does not address the duplicate name error. Option D is wrong because using a different model version does not change the deployment name; the conflict is on the name, not the model version.

Full explanation →

596

Multi-Selectmedium

Which THREE components are required to implement a Retrieval-Augmented Generation (RAG) solution with Azure OpenAI Service? (Choose three.)

Select 3 answers

A.An embedding model (e.g., text-embedding-ada-002)

B.A fine-tuned model

C.An Azure OpenAI Service model (LLM)

D.Azure AI Content Safety

E.A vector database (e.g., Azure AI Search)

AnswersA, C, E

Embedding models convert documents into vector representations.

Why this answer

An embedding model like text-embedding-ada-002 is essential for converting user queries and document chunks into dense vector representations. These vectors enable semantic similarity search in a vector database, which is the core retrieval step in RAG. Without embeddings, the system cannot match user intent to relevant content.

Exam trap

The trap here is that candidates often confuse optional safety or tuning components (like Content Safety or fine-tuning) as mandatory, when the core RAG triad is strictly retrieval (embeddings + vector DB) plus generation (LLM).

Full explanation →

597

MCQmedium

You are deploying a generative AI solution by using Azure OpenAI Service. The solution must ground model responses in a private Azure AI Search index that contains product manuals. You need to configure the model deployment so that the service automatically retrieves relevant chunks from the index and includes them in the prompt at inference time. Which configuration should you use?

A.Upload the manuals to the model's training dataset and enable continuous training on the deployment.

B.Enable the 'Azure OpenAI On Your Data' feature by specifying the Azure AI Search data source in the model's chat completion request.

C.Fine-tune the base model on the contents of the product manuals and deploy the fine-tuned model for inference.

D.Create a custom retrieval pipeline by using Azure Functions that queries the index and appends results to each user message before calling the model.

AnswerB

Azure OpenAI On Your Data lets you attach an Azure AI Search index directly to a chat completion call, and the service handles chunk retrieval, embedding, and prompt assembly server-side, so relevant manual passages are injected before the model generates a response. This is the supported mechanism for grounding without building custom retrieval orchestration.

Why this answer

Azure OpenAI On Your Data is designed to connect a deployment to an Azure AI Search index so that retrieval and prompt augmentation happen automatically at request time. It handles embedding, chunk ranking, and citation generation, which matches the requirement to ground responses in private manuals without custom orchestration.

Exam trap

The trap here is assuming that fine-tuning or uploading documents trains the model to know private content, when runtime grounding actually requires a data source connection such as Azure OpenAI On Your Data.

Full explanation →

598

MCQhard

You are developing a conversational language understanding (CLU) project in Azure AI Language. The project must recognize user intents and extract entities from utterances. You have trained and deployed the model. You need to call the prediction API from a client application. The deployment name is 'prod' and the project name is 'HRBot'. Which URL should you use to send an utterance for prediction?

A.https://<resource-name>.cognitiveservices.azure.com/language/analyze-conversations/projects/HRBot/deployments/prod/:analyze-conversations?api-version=2023-04-01

B.https://<resource-name>.cognitiveservices.azure.com/language/:analyze-conversations?projectName=HRBot&deploymentName=prod&api-version=2023-04-01

C.https://<resource-name>.cognitiveservices.azure.com/language/analyze-text/projects/HRBot/deployments/prod?api-version=2023-04-01

D.https://<resource-name>.cognitiveservices.azure.com/language/analyze-conversations/projects/HRBot/deployments/prod?api-version=2022-05-01

AnswerA

This URL follows the correct REST path for the CLU prediction API: it includes the project name, deployment name, and the analyze-conversations operation with a valid API version. The request body should contain the utterance and optionally the language. This is the endpoint used to get intent and entity predictions from a deployed conversational language understanding model.

Why this answer

The CLU prediction endpoint requires the project name and deployment name as path segments under the analyze-conversations operation, along with a supported API version. The correct URL includes /language/analyze-conversations/projects/{project}/deployments/{deployment}/:analyze-conversations?api-version=2023-04-01. Query-string routing, the wrong operation name, or an outdated API version will not work for CLU predictions.

Exam trap

The trap here is confusing the analyze-text operation used for text analytics with the analyze-conversations operation required for conversational language understanding predictions.

Full explanation →

599

MCQeasy

You need to create a chatbot that uses Azure OpenAI to answer questions about your company's internal policies. The responses must be based only on the provided policy documents. Which approach should you use?

A.Use the model's pre-existing knowledge about common policies.

B.Fine-tune a GPT model on the policy documents.

C.Use prompt engineering to instruct the model to only use policy knowledge.

D.Use Retrieval-Augmented Generation (RAG) with an Azure AI Search index of the documents.

AnswerD

RAG retrieves relevant passages from an Azure AI Search index and grounds Azure OpenAI responses in those documents, satisfying the constraint that answers derive only from the supplied policy content rather than the model's pretrained knowledge.

Why this answer

RAG (Retrieval-Augmented Generation) is the correct approach because it grounds the model's responses in your actual policy documents by retrieving relevant chunks from an Azure AI Search index at query time and injecting them into the prompt. This ensures the chatbot answers only from the provided documents, avoids hallucination, and keeps the source of truth external and updatable without retraining. Azure OpenAI's 'On Your Data' feature implements exactly this pattern using Azure AI Search as the vector/keyword store.

Exam trap

AI-102 often tests the misconception that fine-tuning 'teaches' a model new factual knowledge, when in reality fine-tuning shapes behavior and style while RAG is the correct pattern for grounding responses in specific, updatable documents.

How to eliminate wrong answers

Option A is wrong because the model's pre-existing knowledge is generic, may be outdated, and cannot be verified against your internal policies — it will hallucinate or answer from public data. Option B is wrong because fine-tuning teaches style, format, and tone rather than reliably injecting factual content; the model can still hallucinate, and updating policies would require re-running expensive training jobs. Option C is wrong because prompt engineering alone cannot guarantee the model has access to the actual policy text — without retrieval, the model has no way to know your specific internal documents and will fabricate answers.

Full explanation →

600

Multi-Selecthard

You are developing a custom text classification model using Azure AI Language. You have labeled 2000 documents across 10 categories. You need to evaluate the model's performance before deploying to production. Which THREE metrics should you examine?

Select 3 answers

A.Recall

B.Word Error Rate

C.BLEU Score

D.F1 Score

E.Precision

AnswersA, D, E

Measures the proportion of actual positives correctly identified.

Why this answer

Recall is correct because it measures the proportion of actual positive instances correctly identified by the model, which is critical in custom text classification to ensure that relevant documents are not missed. In Azure AI Language, recall helps assess how well the model captures all instances of each category, especially when class distribution is imbalanced across the 10 categories.

Exam trap

The trap here is that candidates may confuse metrics from other NLP tasks (like speech recognition or translation) with classification metrics, leading them to select Word Error Rate or BLEU Score instead of the standard classification triad of precision, recall, and F1 score.

Full explanation →

Microsoft Azure AI Engineer Associate AI-102 (AI-102) — Questions 526–600