Microsoft Azure AI Engineer Associate AI-102 AI-102 Questions 901–945 | Page 13/13

901

MCQhard

Refer to the exhibit. The indexer using this skillset fails with an error that the skill cannot be executed. The Azure Function is deployed and responds correctly when tested directly. What is the most likely cause?

A.The context is set incorrectly to '/document'.

B.The input source '/document/content' is invalid.

C.The skill endpoint is using HTTP instead of HTTPS.

D.The skill's output targetName does not match the index field.

AnswerC

Azure AI Search requires HTTPS for custom skill endpoints.

Why this answer

Azure AI Search indexers that invoke custom skills via Azure Functions require the endpoint to use HTTPS. If the endpoint is configured with HTTP, the indexer will fail with a 'skill cannot be executed' error, even if the function itself works when tested directly. This is because the indexer enforces secure communication to protect data in transit.

Exam trap

The trap here is that candidates may assume a working function tested directly (often via HTTP) will work in the indexer, overlooking the indexer's strict HTTPS requirement for custom skill endpoints.

How to eliminate wrong answers

Option A is wrong because setting the context to '/document' is standard for document-level skills and does not cause a 'cannot execute' error; it would instead affect how the skill's output is mapped. Option B is wrong because '/document/content' is a valid input source for a document's content field, and if it were invalid, the error would be about missing input data, not skill execution failure. Option D is wrong because a mismatch in output targetName would cause a mapping or indexing error, not a 'skill cannot be executed' error, which occurs before output mapping is evaluated.

Full explanation →

902

MCQhard

Your organization deploys an Azure AI Foundry solution for a customer service chatbot. The chatbot uses a large language model (LLM) hosted on Azure OpenAI Service with a GPT-4 model. Requirements: (1) The chatbot must only use information from the company's internal knowledge base, not general internet knowledge. (2) Responses must include citations from the knowledge base. (3) The solution must filter out any toxic or harmful content. (4) The chatbot must be deployed in a secure environment with network isolation. You have an Azure AI Foundry project with a connected Azure OpenAI resource. The knowledge base is stored in Azure AI Search. You need to configure the solution. What should you do?

A.Use Azure OpenAI with function calling to retrieve knowledge base documents, and enable content filtering.

B.Use prompt engineering with system messages to restrict knowledge, and enable content filtering.

C.Use Azure OpenAI on your data with Azure AI Search as the data source, enable content filtering, and configure the model to use the search index with strict grounding.

D.Fine-tune the GPT-4 model on the knowledge base and deploy with content filtering.

AnswerC

Azure OpenAI on your data grounds responses in the knowledge base, content filtering ensures safety, and strict grounding enforces citations.

Why this answer

It uses Azure OpenAI on your data with Azure AI Search as the data source, which ensures the model only retrieves and generates responses from the indexed knowledge base, meeting the requirement to avoid general internet knowledge. Enabling content filtering satisfies the toxicity requirement, and configuring strict grounding ensures responses include citations from the search index. The secure environment with network isolation is achieved through Azure AI Foundry's managed network capabilities, which are compatible with this configuration.

Exam trap

The trap here is that candidates often confuse fine-tuning (Option D) with RAG, not realizing that fine-tuning cannot provide citations and still risks hallucination, while RAG with Azure AI Search directly satisfies the grounding and citation requirements.

How to eliminate wrong answers

Option A is wrong because function calling allows the model to call external functions but does not restrict the model to only use the knowledge base; it can still generate responses from its training data, violating the requirement to avoid general internet knowledge. Option B is wrong because prompt engineering with system messages is a soft constraint that can be overridden by the model, and it does not guarantee that responses are grounded in the knowledge base or include citations; content filtering alone does not enforce knowledge base usage. Option D is wrong because fine-tuning the GPT-4 model on the knowledge base embeds the data into the model's weights, but it does not provide a mechanism for citing sources, and the model may still hallucinate or use pre-training knowledge; additionally, fine-tuning does not inherently support network isolation or content filtering as a built-in feature.

Full explanation →

903

MCQeasy

Your team is developing a chatbot using Azure AI Bot Service. You need to ensure that the bot can handle multiple languages and respond appropriately. Which Azure AI service should you integrate to perform language detection?

A.Azure AI Language

B.Azure AI Speech

C.Azure AI Content Safety

D.Azure AI Translator

AnswerA

Azure AI Language includes language detection.

Why this answer

Azure AI Language provides pre-built language detection capabilities as part of its natural language processing (NLP) features. By integrating this service, the bot can analyze incoming text and identify the language, enabling it to route responses appropriately or trigger language-specific logic.

Exam trap

The trap here is that candidates often confuse Azure AI Translator's built-in language detection (which is a secondary capability) with the dedicated language detection service, leading them to choose Option D instead of the correct Azure AI Language.

How to eliminate wrong answers

Option B is wrong because Azure AI Speech focuses on speech-to-text, text-to-speech, and speaker recognition, not on detecting the language of text input. Option C is wrong because Azure AI Content Safety is designed to detect harmful or inappropriate content (e.g., hate speech, self-harm) in text or images, not to identify the language. Option D is wrong because Azure AI Translator is used to translate text between languages, but it does not perform standalone language detection; while Translator can sometimes infer language during translation, the dedicated language detection feature is part of Azure AI Language.

Full explanation →

904

MCQhard

A financial services firm wants to use Azure OpenAI to generate investment advice summaries. They must ensure that the model does not produce any advice that could be interpreted as personalized financial advice. What is the most effective strategy?

A.Set temperature to 0 and top_p to 0 to make outputs deterministic.

B.Use a system message that instructs the model to avoid personalized advice and apply strict content filtering.

C.Provide few-shot examples of disclaimers in the prompt.

D.Fine-tune the model on a dataset of generic financial summaries.

AnswerB

System messages and content filtering directly address content restrictions.

Why this answer

Azure OpenAI's system messages allow you to set the model's behavior and constraints at the conversation level, which is the most direct and effective way to enforce a policy like avoiding personalized financial advice. Combined with Azure's content filtering (which can block harmful or restricted content), this approach provides both instruction-based and filter-based guardrails without requiring model retraining or relying solely on example-based prompting.

Exam trap

The trap here is that candidates often assume deterministic parameters (temperature=0, top_p=0) guarantee safe outputs, but they only control randomness, not content compliance—Azure's system message and content filtering are the correct tools for enforcing content policies.

How to eliminate wrong answers

Option A is wrong because setting temperature to 0 and top_p to 0 makes outputs deterministic but does not prevent the model from generating personalized financial advice; it only reduces randomness, not content compliance. Option C is wrong because few-shot examples of disclaimers in the prompt can be ignored or overridden by the model if the underlying training data biases it toward personalized responses; system messages have higher priority in the instruction hierarchy. Option D is wrong because fine-tuning on generic financial summaries requires significant labeled data and compute, and it does not guarantee the model will avoid personalized advice—it may still generate such content if the fine-tuning dataset is not carefully curated to exclude it.

Full explanation →

905

MCQmedium

You are developing a chat application that uses Azure OpenAI GPT-4 to answer customer questions. You need to ensure the model does not generate harmful content. Which configuration should you set?

A.Use a system prompt that instructs the model to be safe.

B.Set the temperature parameter to 0.

C.Set max_tokens to a low value.

D.Enable the content filter in Azure OpenAI Service.

AnswerD

Content filter blocks harmful content.

Why this answer

Azure OpenAI Service includes a built-in content filter that actively scans both input prompts and generated completions to detect and block harmful content such as hate speech, violence, or self-harm. This filter operates at the service level, providing a robust safety layer that cannot be bypassed by model configuration alone. While system prompts can guide behavior, they are not a reliable safeguard against adversarial inputs or model misuse.

Exam trap

The trap here is that candidates assume a system prompt or parameter tuning (temperature, max_tokens) can guarantee safety, but Azure OpenAI's content filter is the only mechanism that actively blocks harmful content at the service level, regardless of model configuration.

How to eliminate wrong answers

Option A is wrong because a system prompt is merely a text instruction and can be overridden by user prompts or jailbreak attempts; it does not enforce content safety at the API or network level. Option B is wrong because setting temperature to 0 only makes the model more deterministic and less creative, but it does not prevent the generation of harmful content if the model's training data includes such patterns. Option C is wrong because max_tokens controls the length of the response, not its safety; a short response can still contain harmful content.

Full explanation →

906

MCQhard

Your organization uses Azure AI Document Intelligence to extract data from invoices. The solution must identify custom fields not present in the prebuilt models, such as 'purchase order number' located in varying positions across documents. What should you do?

A.Use the layout model and apply manual post-processing.

B.Use Azure AI Forms Recognizer with prebuilt receipt model.

C.Use the prebuilt invoice model with field merging.

D.Train a custom extraction model using labeled sample invoices.

AnswerD

Custom models learn to extract user-defined fields.

Why this answer

Azure AI Document Intelligence (formerly Form Recognizer) allows you to train a custom extraction model using labeled sample invoices. This approach enables the model to learn custom fields like 'purchase order number' that appear in varying positions, which prebuilt models cannot handle. By providing labeled examples, the model generalizes to extract the field accurately from new documents.

Exam trap

The trap here is that candidates may assume the prebuilt invoice model can be extended with custom fields via configuration or merging, but Azure AI Document Intelligence requires explicit custom model training to recognize fields not present in prebuilt schemas.

How to eliminate wrong answers

Option A is wrong because the layout model only extracts text and structure (tables, lines) without semantic field recognition; manual post-processing would be inefficient and error-prone for custom fields. Option B is wrong because the prebuilt receipt model is designed for receipts, not invoices, and cannot extract custom fields like 'purchase order number'. Option C is wrong because the prebuilt invoice model does not support field merging; it only extracts predefined fields and cannot learn new custom fields.

Full explanation →

907

MCQhard

A company plans to use Azure Cognitive Search to index a large number of PDF documents stored in Azure Blob Storage. The documents contain sensitive personally identifiable information (PII). The search results should only be accessible to authorized users based on their Azure Active Directory (Azure AD) group membership. Which combination of features should the company implement?

A.Generate shared access signatures (SAS) for each document and include them in the search index.

B.Index the documents and use security filters with Azure AD group-based access control.

C.Use Azure RBAC to assign permissions to users for the search service and configure field-level security.

D.Store Azure AD group membership in a field in the search index and configure an indexer to map the field.

AnswerB

Security filters allow restricting search results based on Azure AD group membership stored in the index.

Why this answer

Azure Cognitive Search supports security trimming via OAuth 2.0 and Azure AD group-based access control. By storing group identifiers in a search index field and using a security filter, only users whose Azure AD group membership matches the stored group IDs can see the corresponding search results. This ensures that sensitive PII in PDF documents is only returned to authorized users without exposing the documents themselves.

Exam trap

The trap here is that candidates confuse Azure RBAC (which controls management-plane permissions) with data-plane security trimming, or they assume that storing group membership in the index alone is sufficient without implementing a query-time filter to enforce it.

How to eliminate wrong answers

Option A is wrong because embedding shared access signatures (SAS) in the search index would expose direct document access URLs to all users who can query the index, bypassing any authorization check; SAS tokens are for delegated access to storage, not for per-user security trimming. Option C is wrong because Azure RBAC controls management-plane access to the search service itself (e.g., who can create indexes or manage the service), not data-plane access to individual search results; field-level security in Cognitive Search is a separate feature that restricts which fields are returned, not which documents are visible based on user identity. Option D is wrong because storing Azure AD group membership in a field alone does not enforce security; the indexer can map the field, but without a security filter applied at query time (e.g., using the `$filter` parameter with the user's group IDs), all documents remain visible to any authenticated user.

Full explanation →

908

MCQhard

A large retail company deploys a custom text classification model using Azure Cognitive Service for Language to categorize customer support tickets into 'Billing', 'Technical', and 'General' categories. The model is trained on 10,000 labeled tickets from the past year. After deployment, the model performs well on new tickets but shows a significant drop in accuracy for tickets submitted during holiday seasons, where the volume of billing issues spikes. The engineering team suspects concept drift. They need to maintain high accuracy without manual retraining every season. Which action should the engineer take?

A.Add subcategories to the classification schema to better capture seasonal patterns.

B.Enable active learning on the model and set up a human review loop for low-confidence predictions.

C.Increase the number of training examples per category to 15,000 each.

D.Schedule monthly retraining using the original 10,000 tickets plus the new tickets.

AnswerB

Active learning continuously improves the model with new data.

Why this answer

Active learning in Azure Cognitive Service for Language automatically identifies low-confidence predictions and sends them for human review, creating a continuous feedback loop that adapts to concept drift without manual retraining. This allows the model to improve its accuracy on seasonal billing spikes by learning from newly labeled examples, while the human review loop ensures quality control.

Exam trap

The trap here is that candidates may think increasing training data or retraining on a schedule is sufficient, but they overlook that active learning with human review is the designed mechanism in Azure Cognitive Service for Language to handle concept drift automatically and continuously.

How to eliminate wrong answers

Option A is wrong because adding subcategories does not address concept drift; it only changes the classification schema without adapting the model to new data distributions. Option C is wrong because simply increasing training examples per category to 15,000 does not incorporate the seasonal billing spike data; the model would still be trained on the original 10,000 tickets and miss the drift. Option D is wrong because scheduling monthly retraining with the original 10,000 tickets plus new tickets is a manual process that does not automatically handle seasonal spikes; it also risks overfitting to recent data and does not leverage the built-in active learning and human review capabilities of the service.

Full explanation →

909

MCQhard

You have the above indexer configuration. The indexer processes a batch of 10 documents. In that batch, 3 documents fail. What happens?

A.The indexer skips the failed documents and continues with the same batch.

B.The indexer stops completely because 3 documents failed.

C.The indexer retries the failed documents.

D.The indexer fails the entire batch but continues with the next batch.

AnswerD

maxFailedItemsPerBatch=2 causes the batch to abort; overall limit of 5 allows subsequent batches.

Why this answer

Azure AI Search indexers use a batch-level failure policy: if the number of failed documents in a batch exceeds the configured 'maxFailedItems' threshold (default 0), the entire batch is marked as failed and skipped, but the indexer continues processing subsequent batches. This behavior is controlled by the indexer's 'maxFailedItems' and 'maxFailedItemsPerBatch' properties, which default to 0, meaning any failure in a batch causes the batch to be skipped while the indexer moves on.

Exam trap

The trap here is that candidates assume individual document failures are silently skipped (Option A) or that any failure stops the entire indexer (Option B), but the actual behavior depends on the configurable 'maxFailedItems' and 'maxFailedItemsPerBatch' thresholds, which cause the batch to fail but allow the indexer to continue if cumulative failures are within limits.

How to eliminate wrong answers

Option A is wrong because the indexer does not skip individual failed documents within a batch when the failure count exceeds the threshold; instead, it fails the entire batch. Option B is wrong because the indexer does not stop completely unless the cumulative failures across all batches exceed the 'maxFailedItems' threshold (default 0), which would stop the entire indexer run, but here only 3 documents fail in one batch, not exceeding the cumulative limit. Option C is wrong because the indexer does not automatically retry failed documents; retry behavior is not part of the default indexer failure handling—failed documents are simply skipped at the batch level.

Full explanation →

910

Multi-Selecteasy

Which TWO monitoring metrics should you track to ensure the health and performance of an Azure AI Search service used for a customer-facing product catalog?

Select 2 answers

A.Throttled search queries count.

B.Indexer execution history and duration.

C.Storage used in GB.

D.Search latency (average and P99).

E.Number of successful search requests.

AnswersA, D

Throttling indicates capacity bottlenecks.

Why this answer

Throttled search queries count (Option A) is a critical health metric because it directly indicates when the search service is under excessive load, causing requests to be rate-limited. For a customer-facing product catalog, throttling degrades user experience and can lead to failed searches. Tracking this metric helps you decide when to scale up replicas or partitions to maintain service level agreements.

Exam trap

The trap here is that candidates often confuse operational metrics (like indexer duration or storage usage) with customer-facing performance metrics, leading them to select indexer execution history instead of search latency.

Full explanation →

911

MCQmedium

You need to implement content moderation for a social media platform using Azure AI Content Safety. The solution must block hate speech and self-harm content while allowing mild profanity. Which configuration should you use?

A.Enable all categories with severity threshold 0 (block all)

B.Enable all categories with severity threshold 2

C.Enable hate speech and self-harm categories; disable profanity

D.Enable hate speech and self-harm with severity threshold 4; enable profanity with severity threshold 6

AnswerD

Threshold 4 blocks high-severity hate/self-harm; threshold 6 allows mild profanity.

Why this answer

Azure AI Content Safety allows you to set severity thresholds per category (0-6). By enabling hate speech and self-harm with a threshold of 4, you block content at severity level 4 or higher (e.g., severe hate speech and self-harm), while enabling profanity with a threshold of 6 means only the most extreme profanity (level 6) is blocked, allowing mild profanity through. This matches the requirement to block hate speech and self-harm but allow mild profanity.

Exam trap

The trap here is that candidates assume disabling a category (like profanity) is the only way to allow it, but Azure AI Content Safety uses severity thresholds to allow mild content while blocking severe content, so disabling the category removes all control over that content type.

How to eliminate wrong answers

Option A is wrong because setting all categories to severity threshold 0 blocks all content, including mild profanity, which violates the requirement to allow mild profanity. Option B is wrong because enabling all categories with severity threshold 2 blocks content at severity level 2 or higher for all categories, including profanity, which would block mild profanity (since mild profanity often falls at severity 2-4). Option C is wrong because disabling the profanity category entirely means no profanity is blocked, but the requirement is to allow mild profanity, not all profanity; disabling the category removes the ability to block any profanity, which could allow severe profanity through.

Full explanation →

912

Multi-Selectmedium

You need to choose Azure services to build a computer vision pipeline that ingests images from multiple sources, extracts text using OCR, and stores extracted metadata in a Cosmos DB database. Which TWO services should you use?

Select 2 answers

A.Azure AI Vision

B.Azure Cognitive Search

C.Azure Functions

D.Azure Logic Apps

E.Azure Blob Storage

AnswersA, C

Provides OCR capabilities.

Why this answer

Azure AI Vision (option A) provides the OCR capability needed to extract text from images via its Read API, which is the core requirement of the pipeline. Azure Functions (option C) is the correct compute service to orchestrate the ingestion, call the OCR API, and write the extracted metadata to Cosmos DB, as it supports event-driven triggers and can scale with image volume. Together, they form a serverless pipeline that meets the stated requirements.

Exam trap

The trap here is that candidates often confuse Azure Cognitive Search as a text extraction service because of its 'cognitive skills' feature, but it is primarily a search indexer that can optionally enrich data, not a direct OCR pipeline component for this specific requirement.

Full explanation →

913

MCQhard

You are analyzing a document using Azure Cognitive Service for Language named entity recognition. The exhibit shows a partial JSON response for entity extraction. The engineer notices that 'Jane Smith' has a low confidence score of 0.45. Which action should the engineer take to improve the confidence score for similar entities?

A.Retrain the entity recognition model with labeled examples of 'Jane Smith'.

B.Provide more context around the entity in the input text, such as titles or roles.

C.Use a different language detection model to improve entity recognition.

D.Decrease the confidence threshold to 0.3 to include 'Jane Smith' as a valid entity.

AnswerB

Additional context helps the model disambiguate.

Why this answer

Providing more context around the entity, such as titles or roles (e.g., 'Dr. Jane Smith' or 'CEO Jane Smith'), gives the prebuilt named entity recognition (NER) model additional linguistic cues that improve its confidence in classifying the entity. Azure Cognitive Service for Language's NER uses a pre-trained model that does not support retraining with custom labels; instead, it relies on surrounding context to disambiguate entities.

Adding descriptive terms helps the model leverage its training on patterns where such context correlates with higher confidence scores.

Exam trap

The trap here is that candidates assume prebuilt NER models can be retrained with labeled examples (Option A), but Azure Cognitive Service for Language's prebuilt NER is a fixed, non-trainable model, and custom retraining requires a separate Custom NER feature.

How to eliminate wrong answers

Option A is wrong because Azure Cognitive Service for Language's prebuilt NER model is not retrainable; custom entity recognition requires a different feature (Custom NER) with a labeled dataset, not the prebuilt API. Option C is wrong because language detection models do not affect entity recognition confidence for a specific entity like a person's name; the language is already detected from the input text, and switching models would not improve entity-level scores. Option D is wrong because decreasing the confidence threshold does not improve the model's confidence score; it only lowers the bar for including entities in results, which may introduce false positives without addressing the underlying recognition quality.

Full explanation →

914

MCQmedium

A retail company uses the Computer Vision Image Analysis API to generate tags for product images in their e-commerce catalog. They want to automatically tag images with product categories such as 'electronics', 'clothing', and 'home goods'. The prebuilt tags often misclassify items. For example, a smartphone is tagged as 'communication device' instead of 'electronics'. You need to improve the tagging accuracy for the company's specific product categories without building a completely new model. What should you do?

A.Train a Custom Vision classification model with images labeled with the company's product categories.

B.Use the Dense Captioning feature to generate detailed descriptions and parse them for categories.

C.Increase the confidence threshold for tags to reduce false positives.

D.Use the 'brands' feature to identify product brands and map them to categories.

AnswerA

Custom Vision can generate custom tags tailored to the company's taxonomy.

Why this answer

Custom Vision allows training a model with custom categories, improving tagging accuracy for specific product categories. Custom Vision works with the Image Analysis API by using the same underlying computer vision capabilities. Option B is incorrect because Dense Captioning provides descriptions but does not allow custom categories.

Option C is incorrect because increasing the confidence threshold only filters out low-confidence tags but does not add custom categories. Option D is incorrect because the brands feature identifies brand names, not product categories.

Full explanation →

915

Multi-Selecthard

Which THREE are required when planning to use Azure OpenAI Service for a generative AI application that must comply with responsible AI principles?

Select 3 answers

A.Restrict the model to a maximum of 1000 tokens.

B.Implement content filters to block harmful outputs.

C.Design with human-in-the-loop for critical decisions.

D.Enable rate limiting to prevent abuse.

E.Establish data governance policies for training data.

AnswersB, C, E

Required for safety.

Why this answer

Azure OpenAI Service provides built-in content filtering that can block harmful outputs such as hate speech, violence, or self-harm. This is a core requirement for responsible AI compliance, as outlined in Microsoft's Responsible AI Standard, and must be configured to prevent the model from generating unsafe content.

Exam trap

The trap here is that candidates confuse operational controls (like token limits or rate limiting) with responsible AI requirements, which are specifically about fairness, safety, transparency, and accountability, not performance or security.

Full explanation →

916

MCQmedium

You are building a chatbot for a retail company using Azure OpenAI Service. The chatbot must provide product recommendations based on customer preferences. To ensure the chatbot does not generate harmful or inappropriate responses, you need to implement a content filtering solution. What should you use?

A.Azure AI Content Safety

B.Azure OpenAI Service content filtering

C.Azure AI Search

D.Microsoft Purview Information Protection

AnswerB

Azure OpenAI Service provides built-in content filtering that can be configured to block harmful content.

Why this answer

Azure OpenAI Service content filtering is the correct choice because it provides built-in, configurable filters that block harmful or inappropriate content at the model level, directly within the Azure OpenAI endpoint. This ensures that the chatbot's product recommendations remain safe without requiring external services, as the filtering is applied to both input prompts and output completions based on severity levels for categories like hate, violence, and self-harm.

Exam trap

The trap here is that candidates often confuse Azure AI Content Safety (a standalone moderation service) with the built-in content filtering of Azure OpenAI Service, assuming that a separate service is required for safety when the native filtering is both sufficient and more tightly integrated.

How to eliminate wrong answers

Option A is wrong because Azure AI Content Safety is a separate service for moderating user-generated content (e.g., images, text) but does not integrate natively with Azure OpenAI Service's model responses; using it would require an extra API call and custom orchestration, adding latency and complexity. Option C is wrong because Azure AI Search is a retrieval service for indexing and querying data (e.g., product catalogs) and has no content filtering capabilities for harmful or inappropriate responses. Option D is wrong because Microsoft Purview Information Protection is a data governance and classification tool for protecting sensitive information (e.g., PII, compliance labels) and does not filter model-generated content for safety or appropriateness.

Full explanation →

917

MCQmedium

Refer to the exhibit. You are configuring an agent in Azure AI Foundry. The agent fails to start because the specified model is not available in the current Azure OpenAI resource. What should you do to resolve the issue?

A.Modify the system_prompt to include the model version

B.Deploy the gpt-4-0613 model in the Azure OpenAI resource

C.Change the connection_type to 'Weak'

D.Change the provider to 'AzureAI'

AnswerB

The model must be deployed before use.

Why this answer

The agent fails to start because the specified model (likely gpt-4-0613) is not deployed in the Azure OpenAI resource. In Azure AI Foundry, agents require an existing model deployment to invoke; you cannot use a model that hasn't been deployed. Option B correctly resolves this by deploying the required model in the Azure OpenAI resource.

Exam trap

The trap here is that candidates might think modifying the system_prompt or changing a connection setting can fix a missing model deployment, but Azure OpenAI requires explicit model deployment before any resource can use it.

How to eliminate wrong answers

Option A is wrong because the system_prompt defines the agent's behavior and instructions, not the model version or deployment; modifying it cannot make an undeployed model available. Option C is wrong because connection_type is not a valid configuration for Azure OpenAI resources; 'Weak' is not a recognized connection type and does not affect model availability. Option D is wrong because the provider is already Azure (Azure OpenAI) and changing it to 'AzureAI' is not a valid provider option; the issue is the missing model deployment, not the provider.

Full explanation →

918

MCQeasy

You are using Azure AI Language Service to extract key phrases from customer reviews. You notice that for reviews containing the word 'not good', the service sometimes extracts 'good' as a key phrase. What is the most likely reason?

A.The language detection model misidentified the language

B.You need to set a confidence threshold to exclude negative phrases

C.Key phrase extraction does not consider negation

D.The service is not trained on your specific domain

AnswerC

Key phrase extraction extracts noun phrases without considering negation modifiers.

Why this answer

Key phrase extraction in Azure AI Language Service uses a statistical model that identifies significant terms based on frequency and context, but it does not inherently understand negation. When the phrase 'not good' appears, the model may still extract 'good' as a key phrase because it recognizes 'good' as a high-value term, ignoring the negation. This is a known limitation of the feature, as it focuses on noun phrases and important terms rather than sentiment or negated constructs.

Exam trap

The trap here is that candidates often assume Azure AI Language Service handles negation across all features, but key phrase extraction explicitly does not consider negation, unlike sentiment analysis which does.

How to eliminate wrong answers

Option A is wrong because language detection is a separate step that identifies the language of the text; misidentification would cause incorrect processing but would not specifically cause 'good' to be extracted from 'not good'. Option B is wrong because confidence thresholds filter out low-confidence phrases, not negative phrases; the service does not have a built-in mechanism to exclude negated terms via threshold settings. Option D is wrong because while domain-specific training can improve accuracy, the core issue here is a fundamental limitation of the key phrase extraction model's handling of negation, not a lack of domain adaptation.

Full explanation →

919

MCQeasy

You are designing an agentic solution that uses Microsoft Copilot Studio and Azure AI Search. The agent needs to answer questions based on confidential documents. Which security measure should you implement to ensure the agent only accesses documents the user has permission to read?

A.Disable public network access on the Azure AI Search service.

B.Implement document-level security using security filters in the search index.

C.Use a managed identity for the agent to access the search index.

D.Require multi-factor authentication for all users.

AnswerB

Security filters enforce permissions at the document level.

Why this answer

Azure AI Search supports document-level security through security filters, which allow you to restrict search results based on the user's identity. By storing security identifiers (e.g., group memberships or user IDs) as a field in the index and applying an OData filter at query time, the agent can ensure users only see documents they are permitted to read. This is the standard approach for implementing row-level security in Azure AI Search.

Exam trap

The trap here is confusing authentication (verifying who the user is) with authorization (determining what the user can access), leading candidates to select network controls or MFA instead of the document-level security filter mechanism.

How to eliminate wrong answers

Option A is wrong because disabling public network access on the Azure AI Search service controls network-level access to the service itself, not document-level permissions within the index; it does not differentiate between users or documents. Option C is wrong because using a managed identity for the agent authenticates the agent to the search service, but does not enforce per-document access control; the agent would have full access to all indexed documents regardless of the end user's permissions. Option D is wrong because requiring multi-factor authentication for all users strengthens authentication but does not restrict which documents a user can see after they are authenticated; it addresses identity verification, not authorization at the document level.

Full explanation →

920

MCQmedium

You are using Azure AI Search to build a knowledge base for a customer support portal. The index includes a 'sentiment' field that should be populated using the Sentiment skill. However, the sentiment scores are not being written to the index. The skillset runs successfully. What is the most likely cause?

A.The output field mapping for 'sentiment' is missing or incorrectly defined in the indexer.

B.The Sentiment skill is not correctly configured in the skillset.

C.The indexer is in a failed state and not processing documents.

D.The sentiment field in the index is of type 'Collection(Edm.String)' but the skill outputs a double.

AnswerA

Without mapping, skill output is not written to index.

Why this answer

The Sentiment skill outputs a 'double' value for sentiment score, but the indexer requires an explicit output field mapping to write that value into the index's 'sentiment' field. Even when a skillset runs successfully, without a correct output field mapping in the indexer definition, the skill's output is not transferred to the index. The indexer's field mappings control how enriched data flows from the skillset's output nodes to the index fields.

Exam trap

The trap here is that candidates assume a successful skillset execution guarantees data is written to the index, but Azure AI Search requires explicit output field mappings in the indexer to bridge skill outputs to index fields, and this step is often overlooked.

How to eliminate wrong answers

Option B is wrong because the question states the skillset runs successfully, meaning the Sentiment skill itself is correctly configured and executed without errors. Option C is wrong because the indexer is explicitly described as running successfully, not in a failed state, so it is processing documents. Option D is wrong because the Sentiment skill outputs a double (a numeric score between 0 and 1), and if the index field were of type 'Collection(Edm.String)', the mismatch would cause an indexer error or warning, but the question says the skillset runs successfully — the issue is the missing mapping, not a type conflict.

Full explanation →

921

MCQmedium

A company is building an agentic solution using Microsoft Copilot Studio. The agent needs to retrieve customer order status from an external CRM API. The API requires OAuth 2.0 authentication with client credentials. Which connector configuration should the developer use?

A.Use a custom connector with API Key authentication.

B.Use the HTTP connector with OAuth2 client credentials grant type.

C.Use a Power Automate flow with a CRM connector that uses service principal.

D.Use an AI Builder model to call the API.

AnswerB

Correct for OAuth 2.0 client credentials flow.

Why this answer

The HTTP connector in Microsoft Copilot Studio supports the OAuth 2.0 client credentials grant type, which is exactly what the external CRM API requires. This grant type allows the agent to authenticate as an application (not a user) by sending a client ID and client secret to obtain an access token, making it ideal for server-to-server API calls where no user interaction is needed.

Exam trap

The trap here is that candidates often confuse the HTTP connector with custom connectors, thinking a custom connector is required for OAuth 2.0, but the HTTP connector natively supports OAuth 2.0 client credentials without needing to build a custom connector.

How to eliminate wrong answers

Option A is wrong because API Key authentication is a simpler, static token method that does not meet the OAuth 2.0 requirement; the CRM API specifically requires OAuth 2.0 with client credentials, not an API key. Option C is wrong because a Power Automate flow with a CRM connector using service principal is an alternative approach but is not a connector configuration within Copilot Studio itself; the question asks for the connector configuration in Copilot Studio, and the HTTP connector is the direct, built-in way to call any REST API with OAuth 2.0 client credentials. Option D is wrong because AI Builder models are designed for AI tasks like prediction or form processing, not for making authenticated API calls to retrieve order status; using AI Builder here would be architecturally incorrect and inefficient.

Full explanation →

922

MCQmedium

You are designing a chatbot using Azure AI Language. The chatbot must understand user intents and also extract entities like dates and locations. Which feature combination should you use?

A.Conversational Language Understanding (CLU) with entities

B.Sentiment analysis and entity linking

C.Custom text classification and key phrase extraction

D.Orchestration Workflow and custom text classification

AnswerA

CLU handles both intents and entities for chatbots.

Why this answer

Conversational Language Understanding (CLU) is the correct Azure AI Language feature for building a chatbot that understands user intents and extracts entities like dates and locations. CLU is specifically designed for natural language understanding (NLU) tasks, providing prebuilt and custom entity extraction alongside intent recognition, which directly matches the requirement.

Exam trap

The trap here is that candidates often confuse entity linking (which maps to external knowledge bases) with entity extraction (which pulls values directly from the utterance), leading them to choose Option B despite it lacking intent recognition.

How to eliminate wrong answers

Option B is wrong because sentiment analysis evaluates the emotional tone of text, not user intents, and entity linking maps named entities to a knowledge base (e.g., Wikipedia), not extracting arbitrary entities like dates and locations. Option C is wrong because custom text classification assigns predefined labels to entire documents, not user intents in a conversational context, and key phrase extraction identifies key terms but does not extract structured entities like dates and locations. Option D is wrong because Orchestration Workflow routes requests between different language services (e.g., CLU, QnA Maker) but does not itself perform intent recognition or entity extraction; custom text classification also does not handle entity extraction.

Full explanation →

923

MCQmedium

A company uses Azure OpenAI to generate code snippets. They notice that the model sometimes produces code that uses deprecated APIs. They want to minimize this without retraining the model. What should they do?

A.Fine-tune the model on a dataset of recent code.

B.Set the temperature parameter to 0 to reduce randomness.

C.Add a system message instructing the model to use only current, non-deprecated APIs.

D.Provide a few-shot example of correct code in the prompt.

AnswerC

System messages effectively guide model behavior.

Why this answer

Adding a system message in Azure OpenAI allows you to set high-level instructions that guide the model's behavior without retraining. By explicitly instructing the model to use only current, non-deprecated APIs, you leverage the system prompt's ability to influence output style and content, effectively reducing deprecated API usage in generated code snippets.

Exam trap

Microsoft often tests the distinction between prompt engineering techniques (system messages, few-shot examples, parameter tuning) and model customization (fine-tuning), and the trap here is that candidates may confuse few-shot prompting (Option D) with a system message, not realizing that a system message provides a more persistent and scalable way to enforce behavioral rules across all responses.

How to eliminate wrong answers

Option A is wrong because fine-tuning requires retraining the model on a custom dataset, which contradicts the requirement to minimize deprecated API usage without retraining. Option B is wrong because setting the temperature parameter to 0 reduces randomness and makes outputs more deterministic, but it does not prevent the model from generating deprecated APIs—it only affects creativity and variability. Option D is wrong because providing a few-shot example of correct code in the prompt can help guide the model, but it is less reliable than a system message for consistently enforcing a rule across all responses, as the model may still default to deprecated APIs in other contexts.

Full explanation →

924

MCQhard

You are reviewing the ARM template snippet for an Azure AI Foundry hub. After deployment, you notice that the hub cannot connect to the storage account. What is the most likely cause?

A.The hub uses a managed network that blocks outbound traffic to the storage account

B.The location 'eastus' does not support AI Foundry hubs

C.The storage account name contains invalid characters

D.The key vault is not configured with a firewall rule

AnswerA

Managed networks can block access to resources without proper firewall rules.

Why this answer

The most likely cause is that the hub uses a managed network which, by default, blocks outbound traffic to the storage account unless explicitly allowed via outbound rules or private endpoints. Azure AI Foundry hubs enforce network isolation for security, and if the storage account is not configured as an approved destination, the hub cannot establish the required connections for data access.

Exam trap

The trap here is that candidates assume connectivity issues are due to resource naming or regional availability, when the real cause is the managed network's default outbound blocking behavior, which is a common security feature in Azure AI services.

How to eliminate wrong answers

Option B is wrong because 'eastus' is a fully supported region for Azure AI Foundry hubs, with no regional restrictions documented. Option C is wrong because storage account names are validated at deployment time; if the name contained invalid characters, the ARM template deployment would fail entirely, not cause a post-deployment connectivity issue. Option D is wrong because the key vault firewall rule is unrelated to the hub's ability to connect to the storage account; key vault is used for secrets, not for storage connectivity.

Full explanation →

925

MCQhard

You are designing a computer vision solution for a retail chain to detect shelf stockouts using store camera feeds. Videos are processed in near real-time. Which combination of Azure services should you use to minimize latency and cost?

A.Use Azure Video Indexer to analyze videos and send results to Azure SQL Database.

B.Use Azure Custom Vision to detect stockouts in video frames.

C.Use Azure Media Services to transcode video and then run Custom Vision on key frames.

D.Use Azure Video Analyzer for Media (formerly Video Indexer) with an Azure IoT Edge module processing video at the edge.

AnswerD

Edge processing reduces latency and bandwidth.

Why this answer

Azure Video Analyzer for Media (formerly Video Indexer) with an Azure IoT Edge module enables near real-time processing at the edge, minimizing latency by avoiding round trips to the cloud, and reduces cost by processing video locally and only sending relevant insights (e.g., stockout detections) to the cloud. This architecture is ideal for retail chains with store camera feeds that require low-latency, cost-effective stockout detection.

Exam trap

The trap here is that candidates often assume cloud-based video analysis (e.g., Video Indexer or Custom Vision) is always the best choice, overlooking the critical need for edge processing to achieve near real-time latency and cost efficiency in distributed retail environments.

How to eliminate wrong answers

Option A is wrong because Azure Video Indexer is designed for media analysis and indexing (e.g., extracting metadata, transcripts) and is not optimized for near real-time stockout detection; sending results to Azure SQL Database adds unnecessary latency and cost. Option B is wrong because Azure Custom Vision alone cannot process video streams in near real-time; it requires frame extraction and lacks built-in video ingestion or edge processing capabilities, leading to high latency and bandwidth costs. Option C is wrong because Azure Media Services is a transcoding and streaming service, not designed for real-time video analysis; transcoding adds latency and cost, and running Custom Vision on key frames still requires cloud round trips, failing to minimize latency.

Full explanation →

926

MCQhard

Your company is building a multi-tenant SaaS application using Azure AI Foundry. Each tenant (customer) should have isolated model deployments and data, but you want to share the base models across tenants to reduce costs. Which approach should you use?

A.Use a single model deployment and a shared endpoint with tenant ID in the header.

B.Create separate fine-tuned models for each tenant.

C.Deploy separate base models for each tenant.

D.Deploy one model per base model and use separate endpoints with routing logic per tenant.

AnswerD

Separate endpoints with routing isolate tenants while sharing the model.

Why this answer

Deploying one model per base model and using separate endpoints with routing logic per tenant provides tenant isolation while sharing the base model, reducing costs. Option A is incorrect because using a single deployment with a tenant ID in the header does not provide sufficient isolation and may lead to data leakage. Option B is incorrect because creating separate fine-tuned models per tenant increases costs and does not share the base model.

Option C is incorrect because deploying separate base models for each tenant also increases costs unnecessarily.

Full explanation →

927

MCQhard

You are building an Azure AI Search solution that indexes data from multiple sources, including SQL Database and Azure Blob Storage. The index must be updated within 15 minutes of any source change. Which approach should you use to achieve near-real-time indexing?

A.Enable incremental enrichment on the skillset

B.Use the push API to send updates as soon as data changes

C.Use an indexer with a schedule set to run every 5 minutes

D.Enable semantic search to speed up indexing

AnswerB

The push API allows you to add or update documents in the index in real-time.

Why this answer

The push API (Azure Cognitive Search REST API or SDK) allows you to directly upload documents to the index as soon as data changes occur, bypassing the indexer's polling cycle. This provides sub-minute latency, meeting the 15-minute near-real-time requirement. Indexers with schedules or enrichment pipelines introduce inherent delays and are not designed for sub-minute updates.

Exam trap

Microsoft often tests the misconception that indexer schedules can achieve near-real-time indexing, but the trap is that indexers have inherent polling intervals and processing overhead that prevent sub-minute latency, making the push API the only viable option for true near-real-time updates.

How to eliminate wrong answers

Option A is wrong because incremental enrichment only optimizes reprocessing of existing documents in a skillset when a skill changes, not the speed of indexing new or updated data from source changes. Option C is wrong because an indexer scheduled every 5 minutes introduces a minimum 5-minute delay plus processing time, which cannot guarantee updates within 15 minutes if the change occurs just after a run. Option D is wrong because semantic search is a query-time feature that improves relevance ranking, not indexing speed or latency.

Full explanation →

928

Multi-Selecthard

Which THREE factors should you consider when selecting a region for an Azure AI Language resource?

Select 3 answers

A.Pricing differences.

B.Availability of the Free tier SKU.

C.Data residency requirements.

D.Language and feature availability.

E.Latency to the application's users.

AnswersC, D, E

Important for compliance.

Why this answer

Data residency requirements (Option C) are critical when selecting an Azure region for an AI Language resource because Azure AI Language processes and stores customer data within the region where the resource is provisioned. Many organizations have compliance obligations (e.g., GDPR, HIPAA, or local data sovereignty laws) that mandate data remain within specific geographic boundaries, so choosing a region that aligns with these requirements is essential.

Exam trap

The trap here is that candidates often confuse operational factors like pricing or SKU availability with the fundamental regional constraints that affect data sovereignty and service capabilities, leading them to select options that are technically true but not the primary factors for region selection in Azure AI Language.

Full explanation →

929

MCQmedium

You are a solution architect at a legal firm. The firm wants to build a copilot using Microsoft Foundry that answers questions about case law documents stored in Azure Blob Storage. The copilot should use the Retrieval Augmented Generation (RAG) pattern with Azure AI Search as the vector store. The documents are in PDF format and include complex tables and footnotes. The solution must ensure that the answers are grounded in the documents and that the copilot can handle follow-up questions. You need to design the ingestion pipeline. Which approach should you take?

A.Use Azure AI Vision OCR to extract text, split by page, and use Azure AI Search keyword search

B.Use Azure AI Document Intelligence prebuilt-read model, chunk by character count, and use Azure AI Search with semantic ranking

C.Use Azure AI Document Intelligence to extract content, then chunk by headings and paragraphs, generate embeddings using Azure OpenAI, and index in Azure AI Search with vector search

D.Use Azure AI Language to extract key phrases, create a non-vector index, and use simple search

AnswerC

Preserves structure and enables RAG with vector search.

Why this answer

It uses Azure AI Document Intelligence to accurately extract content from PDFs (including complex tables and footnotes), then chunks by headings and paragraphs to preserve document structure, generates embeddings via Azure OpenAI for semantic understanding, and indexes in Azure AI Search with vector search to enable RAG-based, grounded answers with follow-up support.

Exam trap

Microsoft often tests the misconception that simple OCR or keyword search is sufficient for complex documents, but the trap here is that legal documents with tables and footnotes require structure-aware extraction and vector search to support grounded, conversational RAG.

How to eliminate wrong answers

Option A is wrong because Azure AI Vision OCR is designed for image-based text extraction and lacks the ability to handle complex tables and footnotes in PDFs; splitting by page ignores document structure, and keyword search alone cannot support semantic understanding or follow-up questions. Option B is wrong because the prebuilt-read model extracts raw text without preserving table/footnote structure, chunking by character count breaks logical content boundaries, and semantic ranking on keyword search does not provide the vector-based retrieval needed for RAG. Option D is wrong because key phrase extraction loses document context and structure, a non-vector index cannot support semantic similarity search, and simple search cannot ground answers in document content or handle follow-up questions effectively.

Full explanation →

930

Multi-Selecthard

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

Select 3 answers

A.A bot channel registration or web app to interact with users

B.An Azure AI Language Service resource

C.A set of FAQ documents or web pages as the knowledge source

D.An Azure AI Search index

E.An Azure Functions app for serverless compute

AnswersA, B, C

A client interface is needed for users to submit questions.

Why this answer

A bot channel registration or web app is the front-end interface that allows users to submit questions and receive answers from the custom QnA solution. Without this interaction layer, the knowledge base cannot be queried by end users, making it an essential component for a complete solution.

Exam trap

The trap here is that candidates often assume Azure AI Search is mandatory because it is used in other Azure AI services like Cognitive Search, but for custom question answering the search index is automatically provisioned and managed by the Language Service itself.

Full explanation →

931

MCQeasy

A developer is tasked with integrating Azure OpenAI Service into an application that generates product descriptions. The developer needs to ensure that the generated content does not contain offensive language. Which Azure AI service should be used in addition to Azure OpenAI?

A.Azure AI Search

B.Azure AI Vision

C.Azure AI Language

D.Azure AI Content Safety

AnswerD

Azure AI Content Safety detects offensive language and other harmful content.

Why this answer

Azure AI Content Safety (D) is the correct service because it provides built-in content moderation capabilities that can detect and filter offensive, inappropriate, or harmful language in text and images. By integrating Azure AI Content Safety with Azure OpenAI, the developer can automatically screen generated product descriptions for profanity, hate speech, or other offensive content before they are displayed to users, ensuring compliance with content policies.

Exam trap

The trap here is that candidates may confuse Azure AI Language's text analytics features (like sentiment analysis) with content moderation, but Azure AI Language does not include dedicated offensive language filtering, which is a distinct capability of Azure AI Content Safety.

How to eliminate wrong answers

Option A is wrong because Azure AI Search is a cognitive search service used for indexing and retrieving data, not for content moderation or filtering offensive language. Option B is wrong because Azure AI Vision is designed for image analysis tasks such as object detection, OCR, and facial recognition, and does not include text-based content safety features. Option C is wrong because Azure AI Language provides natural language processing capabilities like sentiment analysis, key phrase extraction, and language understanding, but it does not offer dedicated content moderation or offensive language detection; that functionality is specifically handled by Azure AI Content Safety.

Full explanation →

932

Multi-Selecteasy

You are using Azure AI Language to analyze social media comments. You need to identify the language of each comment and then extract key phrases. Which TWO features should you use? (Select TWO.)

Select 2 answers

A.Sentiment analysis

B.Summarization

C.Language detection

D.Entity recognition

E.Key phrase extraction

AnswersC, E

Language detection identifies the language of the text.

Why this answer

Language detection is the correct feature because it identifies the language of each comment, which is a prerequisite for further analysis. Key phrase extraction is the second correct feature because it extracts important terms from the text, directly addressing the requirement to 'extract key phrases' after language identification.

Exam trap

Microsoft Azure AI Language often tests the distinction between features that analyze content (sentiment, entities, key phrases) versus those that identify metadata (language), and the trap here is that candidates might confuse 'key phrase extraction' with 'entity recognition' because both extract terms, but key phrases are broader and not limited to named entities.

Full explanation →

933

MCQmedium

Your organization is using Azure AI Search to index a large collection of PDF documents stored in Azure Blob Storage. The index currently returns search results, but users complain that the results are not relevant when they search using natural language phrases. You need to improve the relevance of search results without rewriting the application. What should you do?

A.Increase the number of replicas for the search service to improve query performance.

B.Create a new index with a blob indexer that uses the 'content' field only.

C.Enable semantic search on the index and configure a semantic configuration.

D.Configure a custom analyzer on the index to handle stop words and synonyms.

AnswerC

Semantic search uses AI models to improve relevance of natural language queries.

Why this answer

Semantic search in Azure AI Search uses advanced language models to understand the intent behind natural language queries, re-ranking results based on semantic relevance rather than just keyword matching. Enabling semantic search and configuring a semantic configuration directly addresses the user complaint about poor relevance for natural language phrases without requiring application changes.

Exam trap

The trap here is that candidates often confuse improving query performance (replicas) or basic text processing (custom analyzers) with the semantic understanding needed for natural language queries, leading them to pick options that address performance or tokenization rather than relevance.

How to eliminate wrong answers

Option A is wrong because increasing replicas only improves query throughput and availability, not the relevance or semantic understanding of search results. Option B is wrong because creating a new index with only the 'content' field would reduce the available data for matching, likely worsening relevance rather than improving it. Option D is wrong because custom analyzers handle tokenization, stop words, and synonyms at indexing time, but they do not provide the deep semantic understanding needed to interpret natural language phrases; semantic search is required for that.

Full explanation →

934

Multi-Selectmedium

Which TWO Azure AI services can be used to build a multilingual question-answering bot that retrieves answers from a knowledge base of documents?

Select 2 answers

A.Azure AI Language Understanding (LUIS)

B.Azure OpenAI Service with a RAG pattern

C.Azure AI Translator

D.Azure AI Document Intelligence

E.Azure AI Language - Custom Question Answering

AnswersB, E

Azure OpenAI with RAG can retrieve answers from documents using embeddings.

Why this answer

The Azure OpenAI Service combined with a Retrieval-Augmented Generation (RAG) pattern allows you to ingest documents into a vector store, retrieve relevant chunks based on the user's query, and then generate a natural language answer using a large language model. This architecture inherently supports multilingual input and output because the underlying model (e.g., GPT-4) can process and generate text in many languages, making it ideal for a multilingual question-answering bot.

Exam trap

The AI-102 exam often tests the distinction between services that process language (like LUIS or Translator) versus services that combine retrieval with generation (like Azure OpenAI with RAG) to answer questions from documents, leading candidates to mistakenly choose LUIS or Translator for a task that requires document-based Q&A.

Full explanation →

935

MCQhard

You are designing a generative AI solution that uses Azure OpenAI Service. The solution must generate code snippets in Python and JavaScript. You need to ensure the model reliably outputs code in the correct language based on user input. Which approach should you use?

A.Set the top_p parameter to a low value.

B.Use a system message to specify the desired language.

C.Fine-tune the model on a dataset of code in both languages.

D.Set the temperature to 0 to make the model deterministic.

AnswerB

System messages can instruct the model on the language to use.

Why this answer

System messages in Azure OpenAI Service allow you to set the context or behavior of the model, such as specifying the desired programming language for code generation. This approach is lightweight, requires no retraining, and reliably guides the model to output code in the correct language based on the user's request, leveraging the model's existing training on both Python and JavaScript.

Exam trap

The trap here is that candidates often confuse hyperparameters like temperature and top_p with content control mechanisms, mistakenly believing they can enforce output language, when in fact they only affect randomness and token selection probability.

How to eliminate wrong answers

Option A is wrong because setting top_p to a low value reduces the pool of tokens considered for sampling, which can make outputs more focused but does not control the language of the generated code; it is a nucleus sampling parameter, not a language selector. Option C is wrong because fine-tuning the model on a dataset of code in both languages is overkill for this requirement, as the base model already understands both languages; fine-tuning is typically used for specialized tasks or to adapt to a specific domain, not for simple language switching. Option D is wrong because setting temperature to 0 makes the model deterministic by always choosing the most likely token, but it does not enforce the output language; it can still produce code in the wrong language if the prompt is ambiguous, and it reduces creativity but does not guarantee language adherence.

Full explanation →

936

MCQmedium

You are deploying a generative AI application. You have the JSON configuration above. You need to ensure that the model stops generating output at the end of each sentence. The current configuration uses a stop sequence of "\n". What should you do?

A.Change the stop sequence to [".", "!" , "?"]

B.Reduce the maxTokens to 500.

C.Set the topP to 0.5.

D.Set the temperature to 0.

AnswerA

These are sentence-ending punctuation marks.

Why this answer

The goal is to stop generation at the end of each sentence. A sentence can end with a period (.), exclamation mark (!), or question mark (?). The current stop sequence of "\n" only stops at newline characters, which does not guarantee sentence boundaries.

By changing the stop sequence to an array containing [".", "!", "?"], the model will halt output whenever it generates any of these punctuation marks, ensuring each sentence is complete.

Exam trap

The trap here is that candidates often confuse parameters that control output randomness (temperature, topP) or length (maxTokens) with parameters that define explicit stopping conditions, leading them to select options that do not actually enforce sentence boundaries.

How to eliminate wrong answers

Option B is wrong because reducing maxTokens to 500 limits the total number of tokens generated but does not control where generation stops relative to sentence boundaries; the model could still cut off mid-sentence. Option C is wrong because setting topP to 0.5 adjusts the nucleus sampling threshold, affecting token selection diversity, not the stopping condition; it does not enforce sentence-ending punctuation. Option D is wrong because setting temperature to 0 makes the model deterministic (greedy decoding) but does not introduce a stop condition based on sentence-ending characters; the model will still generate until maxTokens or another stop sequence is reached.

Full explanation →

937

MCQmedium

You are building a chatbot that must handle customer inquiries about order status. The solution must use Azure AI Language and support multiple languages. You need to configure the project to detect language automatically from user input. Which setting should you enable?

A.Set the primary language to English only.

B.Use the conversation summarization feature.

C.Enable multi-lingual detection in the project language settings.

D.Enable spelling correction in the project settings.

AnswerC

Allows the project to automatically detect the language of each utterance.

Why this answer

Azure AI Language's multi-lingual detection setting enables the project to automatically identify the language of user input without requiring explicit language selection. This is essential for a chatbot handling customer inquiries in multiple languages, as it allows the service to route the input to the appropriate language-specific model or processing pipeline, ensuring accurate order status responses across languages.

Exam trap

The trap here is that candidates confuse 'multi-lingual detection' with 'spelling correction' or 'conversation summarization', mistakenly thinking those features handle language identification, when in fact they address different aspects of text processing.

How to eliminate wrong answers

Option A is wrong because setting the primary language to English only would force all input to be processed as English, defeating the requirement to support multiple languages and automatically detect language from user input. Option B is wrong because conversation summarization is a feature that generates summaries of conversations, not a language detection mechanism; it does not enable automatic language identification. Option D is wrong because spelling correction improves input accuracy by fixing typos but does not detect or handle language detection; it operates on the text after language is already determined.

Full explanation →

938

Multi-Selectmedium

Which TWO Azure AI services can be used to extract text from images and PDFs? (Select two.)

Select 2 answers

A.Azure AI Translator

B.Azure AI Search

C.Azure AI Vision OCR

D.Azure AI Document Intelligence

E.Azure AI Language

AnswersC, D

OCR extracts text from images and PDFs.

Why this answer

Azure AI Vision OCR (Optical Character Recognition) is designed to extract printed and handwritten text from images, while Azure AI Document Intelligence (formerly Form Recognizer) specializes in extracting text, tables, and key-value pairs from documents, including PDFs. Both services provide robust text extraction capabilities from visual sources, making them the correct choices.

Exam trap

The trap here is that candidates may confuse Azure AI Language's text analysis capabilities with OCR, or assume Azure AI Search can extract text directly, when in fact it only indexes pre-extracted data.

Full explanation →

939

MCQeasy

A company is deploying an Azure AI solution that uses Azure Cognitive Services. The solution must comply with data residency requirements that mandate all customer data be stored within a specific geographic region. Which action should the company take when creating the Cognitive Services resource?

A.Apply a resource tag that specifies the region.

B.Configure the endpoint URL to point to a regional endpoint.

C.Set the SKU to a tier that supports regional restrictions.

D.Select the appropriate region during resource creation.

AnswerD

The resource location determines where data is stored at rest.

Why this answer

Data residency requirements are satisfied by physically storing customer data within a specific geographic boundary. When creating an Azure Cognitive Services resource, selecting the appropriate region (e.g., 'West Europe' or 'East US') during the provisioning process ensures that all data processed and stored by that service instance remains within that Azure datacenter region. This is the fundamental and only guaranteed method to enforce data residency at the resource level.

Exam trap

The trap here is that candidates confuse network-level controls (like endpoint configuration or tagging) with physical data storage guarantees, mistakenly believing that a regional endpoint or a tag can enforce data residency when only the initial region selection during resource creation can do so.

How to eliminate wrong answers

Option A is wrong because resource tags are metadata labels used for organization, cost tracking, or policy enforcement; they do not influence where the underlying service stores data. Option B is wrong because the endpoint URL is automatically generated based on the chosen region and cannot be manually configured to redirect storage; it only determines the network access point, not the physical data location. Option C is wrong because the SKU tier (e.g., S0, F0) determines throughput limits and feature availability, not geographic restrictions; no SKU tier enforces regional data storage.

Full explanation →

940

MCQmedium

Your organization uses Microsoft Purview to govern AI models. You need to ensure that data used for training complies with internal policies. What should you configure?

A.Set up Data Estate Insights to monitor data usage

B.Use Microsoft Purview Data Map to catalog training data

C.Apply sensitivity labels to training data

D.Enable Microsoft Purview Information Protection

AnswerA

Provides governance and compliance monitoring.

Why this answer

Data Estate Insights in Microsoft Purview provides monitoring and analytics over data estate health, including data usage patterns. By configuring Data Estate Insights, you can track how training data is accessed and used, enabling you to verify compliance with internal policies on data consumption for AI model training.

Exam trap

The trap here is confusing data cataloging (Data Map) or classification (sensitivity labels) with usage monitoring, leading candidates to select options that address data discovery or protection rather than compliance monitoring.

How to eliminate wrong answers

Option B is wrong because Microsoft Purview Data Map is used for metadata scanning, classification, and lineage tracking of data assets, not for monitoring data usage compliance. Option C is wrong because applying sensitivity labels classifies data based on sensitivity but does not monitor or enforce usage policies for training data. Option D is wrong because Microsoft Purview Information Protection focuses on data protection through encryption and access controls, not on monitoring data usage for compliance.

Full explanation →

941

MCQmedium

You have an Azure AI solution that uses Azure AI Language to perform sentiment analysis. The solution is experiencing high latency. Which action should you take to reduce latency?

A.Move the service to a different Azure region.

B.Use the Free tier of the Azure AI Language service.

C.Increase the request timeout value.

D.Scale the service by increasing the number of instances or using a higher pricing tier.

AnswerD

More capacity reduces queuing and latency.

Why this answer

Scaling the Azure AI Language service by increasing the number of instances or moving to a higher pricing tier (e.g., from Standard S0 to a tier with higher throughput) directly addresses high latency by providing more capacity to handle concurrent requests. High latency often results from hitting the service's rate limits or throughput constraints, and scaling alleviates this bottleneck without changing the underlying architecture.

Exam trap

The trap here is that candidates often confuse network latency (solved by region proximity) with service throughput latency (solved by scaling), leading them to incorrectly choose Option A when the real bottleneck is capacity, not geography.

How to eliminate wrong answers

Option A is wrong because moving the service to a different Azure region primarily reduces network latency due to geographic proximity, but it does not resolve high latency caused by insufficient service capacity or throttling; the core issue is throughput, not distance. Option B is wrong because the Free tier has strict rate limits (e.g., 5,000 transactions per month) and lower throughput, which would likely worsen latency under load rather than reduce it. Option C is wrong because increasing the request timeout value does not reduce latency; it only allows the client to wait longer for a response, masking the symptom without addressing the underlying performance issue.

Full explanation →

942

MCQhard

You are deploying an Azure AI multi-agent solution on Microsoft Foundry. The solution uses three agents that must share context and pass tasks among themselves. To ensure reliable orchestration and traceability, which agent pattern should you implement?

A.Parallel agent pattern

B.Sequential agent pattern

C.Sub-agent pattern

D.Custom agent pattern

AnswerB

Sequential pattern ensures ordered task execution with clear handoffs and traceability.

Why this answer

The sequential agent pattern is correct because the scenario requires agents to share context and pass tasks among themselves in a reliable, traceable manner. In Microsoft Foundry, the sequential pattern ensures each agent executes in a defined order, passing state and results to the next agent, which provides clear orchestration and full traceability of the workflow.

Exam trap

The trap here is that candidates often confuse the sequential pattern with the parallel pattern, assuming that 'sharing context' implies concurrent execution, but the question's emphasis on 'reliable orchestration and traceability' explicitly requires ordered, non-concurrent task handoff.

How to eliminate wrong answers

Option A is wrong because the parallel agent pattern executes agents concurrently, which does not guarantee ordered task passing or shared context across agents in a sequential dependency. Option C is wrong because the sub-agent pattern is used for delegating subtasks to a child agent from a parent, not for peer-to-peer context sharing and task handoff among multiple agents. Option D is wrong because the custom agent pattern is a generic term for user-defined agent logic, not a specific orchestration pattern that ensures reliable sequential task passing and traceability.

Full explanation →

943

MCQeasy

You need to analyze the sentiment of social media posts in real time using Azure AI Language. Which approach should you use?

A.Call the sentiment analysis REST API for each post

B.Use Azure AI Search with cognitive skills

C.Use the batch processing feature in Azure AI Language

D.Send posts to an Event Hub and use Stream Analytics

AnswerA

The API provides real-time analysis.

Why this answer

The sentiment analysis REST API in Azure AI Language is designed for real-time, per-document analysis. By calling the API for each social media post as it arrives, you achieve the lowest latency and can process posts individually without batching or streaming overhead, which is essential for real-time sentiment analysis.

Exam trap

The trap here is that candidates often confuse real-time processing with streaming architectures (like Event Hubs and Stream Analytics) or batch processing, but the simplest and most direct real-time approach for per-document sentiment analysis is the REST API.

How to eliminate wrong answers

Option B is wrong because Azure AI Search with cognitive skills is designed for indexing and enriching documents at rest, not for real-time processing of individual streaming posts. Option C is wrong because the batch processing feature in Azure AI Language is intended for asynchronous, high-throughput processing of large volumes of documents, not for real-time, per-post analysis. Option D is wrong because sending posts to an Event Hub and using Stream Analytics is a streaming architecture that adds unnecessary complexity and latency for simple per-post sentiment analysis; the REST API is more direct and efficient for real-time needs.

Full explanation →

944

MCQmedium

You have the above Azure AI Search skillset. The indexer fails with the error 'The skill 'sentiment-skill' cannot find the input '/document/pages/*' because the path does not exist.' What is the most likely cause?

A.The SplitSkill did not split the content because the document content is too short.

B.The SplitSkill is not defined in the skillset.

C.The SentimentSkill is missing a required input.

D.The targetName in SplitSkill is misspelled.

AnswerA

If content is too short, no pages are produced.

Why this answer

The error indicates that the input path '/document/pages/*' does not exist. This path is generated by the SplitSkill when it splits the document content into pages. If the SplitSkill does not produce any output, the path will not exist.

The most likely cause is that the document content is too short to be split, so no pages are created. Therefore, option A is correct. Option B is incorrect because the SplitSkill is defined.

Option C is incorrect because the SentimentSkill's input path is correct, but the source is missing. Option D is incorrect because the error is about a missing path, not a misspelled targetName.

Full explanation →

945

MCQhard

You are a senior AI engineer at a financial services company. You are building a generative AI solution to assist financial advisors with client portfolio recommendations. The solution must use Azure OpenAI Service. The following requirements must be met: 1. Responses must be based on the latest market data and client profiles stored in Azure SQL Database. 2. The solution must not generate investment advice that is not backed by the data. 3. The solution must be cost-effective and minimize API calls. 4. The system must provide citations for the data used in the response. You design a RAG pattern with Azure AI Search indexing the portfolio data. You also implement a system message instructing the model to only use provided context. However, the model occasionally generates advice that contradicts the data or invents new facts. You need to modify the solution to ensure responses are strictly grounded in the retrieved data. What should you do?

A.Set temperature to 0 in the Azure OpenAI completion request.

B.Reduce the top_k parameter in the search query to retrieve fewer documents.

C.Increase the chunk size in the index to provide more context per document.

D.Use Azure OpenAI Service on your own data integration to directly query the SQL database.

AnswerA

Temperature 0 makes output deterministic and grounded in provided context.

Why this answer

Setting the temperature to 0 makes the model deterministic, reducing its tendency to generate creative or ungrounded responses. This enforces adherence to the provided context (retrieved data) and minimizes hallucination. Option B is incorrect: reducing top_k limits the number of retrieved documents but does not force the model to rely solely on them; the model can still invent facts if the temperature is high.

Option C is incorrect: increasing chunk size provides more context per document but does not prevent the model from ignoring or contradicting that context. Option D is incorrect: Azure OpenAI Service on your own data still uses a language model that can hallucinate if not properly constrained (e.g., with low temperature); it does not directly query SQL databases in real time.

Full explanation →

Microsoft Azure AI Engineer Associate AI-102 (AI-102) — Questions 901–945