AI-102 Study Guide — 60 Practice Questions

Q: You are designing an Azure AI solution that needs to process handwritten documents from multiple languages. Which Azure service should you primarily use to extract text from these images?

Correct answer: Azure Computer Vision API with OCR — Azure Computer Vision API with OCR capabilities is specifically designed to extract text from images including handwritten content. While Form Recognizer is also useful, Computer Vision is the primary service for general OCR tasks across multiple languages.

Q: Your organization wants to build a chatbot that understands user intent and extracts entities from conversations. Which combination of Azure Cognitive Services would be most appropriate?

Correct answer: Conversational Language Understanding and Entity Recognition — Conversational Language Understanding (the modern successor to LUIS) combined with entity recognition capabilities provides the best foundation for building intent-driven chatbots. These services work together to understand what users want to do and extract relevant information.

Q: You need to implement a solution that detects anomalies in time-series data from IoT sensors. What should you use?

Correct answer: Azure Anomaly Detector — Azure Anomaly Detector is purpose-built for detecting anomalies in time-series data and includes pre-built models optimized for sensor data analysis. It provides both univariate and multivariate anomaly detection capabilities.

Q: You are implementing a document classification solution using Azure Form Recognizer. You need to classify invoices from different vendors with varying formats. What is the recommended approach?

Correct answer: Train custom models for each vendor format using labeled samples — Training custom models with labeled samples allows Form Recognizer to learn vendor-specific layouts and formats. This supervised approach provides better accuracy than using only prebuilt models or relying on downstream text analysis.

Q: Your solution requires real-time translation of user input across 50+ languages. Which approach minimizes latency while ensuring accuracy?

Correct answer: Cache translations using Azure Cache for Redis with Translator API — Using Azure Cache for Redis alongside the Translator API minimizes latency by caching frequently translated phrases while the Translator API ensures accuracy across all 50+ languages without the overhead of local model deployment.

Q: You need to build a solution that analyzes customer sentiment from social media posts in real-time. What is the primary service you should implement?

Correct answer: Azure Text Analytics with sentiment analysis — Azure Text Analytics includes sentiment analysis capabilities that can analyze emotional tone in text, making it ideal for social media monitoring. It can process multiple languages and provide confidence scores for sentiment classification.

Q: You are designing an AI solution that must comply with responsible AI principles and needs to audit model decisions. Which Azure service provides explainability and interpretability features?

Correct answer: Azure Machine Learning Responsible AI dashboard — The Azure Machine Learning Responsible AI dashboard provides tools for model explainability, fairness assessment, and interpretability. It helps audit and understand how models make decisions, which is essential for responsible AI implementation.

Q: Your organization needs to extract key information from purchase orders with consistent structure but from different suppliers. What Form Recognizer feature should you use?

Correct answer: Custom model trained with labeled examples — Since the documents have consistent structure but come from different suppliers with potentially different formats, a custom model trained with labeled examples from your specific suppliers will provide the best accuracy for extracting your required fields.

Q: You are implementing a multi-modal AI solution that processes both images and text together. Which architecture pattern would be most appropriate?

Correct answer: Parallel processing of image and text features with fusion — Parallel processing of image and text features with fusion allows the model to leverage both modalities simultaneously, capturing relationships between visual and textual information that sequential processing might miss.

Q: You need to implement object detection in images to identify specific products in retail environments. Which Azure service would you choose?

Correct answer: Azure Computer Vision with object detection model — Azure Computer Vision includes object detection capabilities that can identify and locate objects in images. For retail product detection at scale, the prebuilt models are efficient, though Custom Vision could be used if you need custom product categories.

Q1 Medium

You are designing an Azure AI solution that needs to process handwritten documents from multiple languages. Which Azure service should you primarily use to extract text from these images?

A Azure Computer Vision API with OCR ✓ Correct
B Azure Form Recognizer for document analysis
C Azure Cognitive Search for document indexing
D Azure Text Analytics for language detection

Explanation

Azure Computer Vision API with OCR capabilities is specifically designed to extract text from images including handwritten content. While Form Recognizer is also useful, Computer Vision is the primary service for general OCR tasks across multiple languages.

Q2 Medium

Your organization wants to build a chatbot that understands user intent and extracts entities from conversations. Which combination of Azure Cognitive Services would be most appropriate?

A Conversational Language Understanding and Entity Recognition ✓ Correct
B Speech Recognition and Form Recognizer
C Language Understanding (LUIS) and Text Analytics
D QnA Maker and Translator

Explanation

Conversational Language Understanding (the modern successor to LUIS) combined with entity recognition capabilities provides the best foundation for building intent-driven chatbots. These services work together to understand what users want to do and extract relevant information.

Q3 Easy

You need to implement a solution that detects anomalies in time-series data from IoT sensors. What should you use?

A Azure Anomaly Detector ✓ Correct
B Azure Stream Analytics with custom functions
C Azure Synapse Analytics for pattern detection
D Azure Machine Learning with AutoML

Explanation

Azure Anomaly Detector is purpose-built for detecting anomalies in time-series data and includes pre-built models optimized for sensor data analysis. It provides both univariate and multivariate anomaly detection capabilities.

Q4 Medium

You are implementing a document classification solution using Azure Form Recognizer. You need to classify invoices from different vendors with varying formats. What is the recommended approach?

A Train custom models for each vendor format using labeled samples ✓ Correct
B Use Optical Character Recognition followed by Text Analytics classification
C Implement a custom machine learning pipeline in Azure ML
D Use prebuilt invoice model only for all vendors

Explanation

Training custom models with labeled samples allows Form Recognizer to learn vendor-specific layouts and formats. This supervised approach provides better accuracy than using only prebuilt models or relying on downstream text analysis.

Q5 Hard

Your solution requires real-time translation of user input across 50+ languages. Which approach minimizes latency while ensuring accuracy?

A Call Azure Translator API directly for each language pair
B Cache translations using Azure Cache for Redis with Translator API ✓ Correct
C Implement local language models using Azure Machine Learning
D Use Azure Translator with batch processing mode

Explanation

Using Azure Cache for Redis alongside the Translator API minimizes latency by caching frequently translated phrases while the Translator API ensures accuracy across all 50+ languages without the overhead of local model deployment.

Q6 Easy

You need to build a solution that analyzes customer sentiment from social media posts in real-time. What is the primary service you should implement?

A Azure Text Analytics with sentiment analysis ✓ Correct
B Azure Language Understanding for intent detection
C Azure Cognitive Search for content indexing
D Azure Bot Service for conversation management

Explanation

Azure Text Analytics includes sentiment analysis capabilities that can analyze emotional tone in text, making it ideal for social media monitoring. It can process multiple languages and provide confidence scores for sentiment classification.

Q7 Hard

You are designing an AI solution that must comply with responsible AI principles and needs to audit model decisions. Which Azure service provides explainability and interpretability features?

A Azure Machine Learning Responsible AI dashboard ✓ Correct
B Azure Policy for compliance monitoring
C Azure Cognitive Services audit logs
D Azure Security Center for model protection

Explanation

The Azure Machine Learning Responsible AI dashboard provides tools for model explainability, fairness assessment, and interpretability. It helps audit and understand how models make decisions, which is essential for responsible AI implementation.

Q8 Medium

Your organization needs to extract key information from purchase orders with consistent structure but from different suppliers. What Form Recognizer feature should you use?

A Prebuilt invoice model
B Layout analysis without field extraction
C General document analysis with OCR
D Custom model trained with labeled examples ✓ Correct

Explanation

Since the documents have consistent structure but come from different suppliers with potentially different formats, a custom model trained with labeled examples from your specific suppliers will provide the best accuracy for extracting your required fields.

Q9 Hard

You are implementing a multi-modal AI solution that processes both images and text together. Which architecture pattern would be most appropriate?

A Image-only processing with metadata extraction
B Sequential processing: OCR first, then text analysis
C Parallel processing of image and text features with fusion ✓ Correct
D Text-only processing after OCR conversion

Explanation

Parallel processing of image and text features with fusion allows the model to leverage both modalities simultaneously, capturing relationships between visual and textual information that sequential processing might miss.

Q10 Medium

You need to implement object detection in images to identify specific products in retail environments. Which Azure service would you choose?

A Azure Form Recognizer for document objects
B Azure Computer Vision with object detection model ✓ Correct
C Azure Video Indexer for video analysis
D Azure Custom Vision for trained object detection

Explanation

Azure Computer Vision includes object detection capabilities that can identify and locate objects in images. For retail product detection at scale, the prebuilt models are efficient, though Custom Vision could be used if you need custom product categories.

Q11 Medium

Your solution requires processing confidential customer documents. What security consideration is most critical when implementing Azure AI services?

A Storing API keys in configuration files with restricted access
B Implementing end-to-end encryption and using private endpoints ✓ Correct
C Disabling all logging and monitoring features
D Using only free tier services to minimize exposure

Explanation

End-to-end encryption and private endpoints ensure that sensitive data is protected both in transit and at rest when communicating with Azure AI services. This is essential for processing confidential customer information.

Q12 Hard

You are building a recommendation system using Azure Machine Learning. Which approach would handle cold-start problems for new users most effectively?

A Hybrid approach combining content-based and collaborative filtering ✓ Correct
B Collaborative filtering exclusively
C Random recommendations until sufficient user history exists
D Content-based filtering using item features

Explanation

A hybrid approach combines content-based filtering (which works without user history) with collaborative filtering (which improves with data), effectively addressing the cold-start problem for new users while maintaining recommendation quality.

Q13 Medium

You need to monitor the performance of a deployed Azure AI model and detect when it starts underperforming. What should you implement?

A Manual testing after each deployment
B Application Insights with custom metrics and alerts ✓ Correct
C Only track inference count and API response times
D Compare predictions against training data accuracy

Explanation

Application Insights allows you to set up custom metrics and alerts to monitor model performance, including accuracy metrics and data drift detection. This enables proactive detection of degradation in production models.

Q14 Medium

Your organization wants to implement a knowledge base solution that automatically extracts Q&A pairs from existing documentation. Which service combination would be optimal?

A Text Analytics with Azure Search indexing
B Azure Bot Service with Cognitive Search
C Language Understanding with manual knowledge entry
D QnA Maker with Form Recognizer for document extraction ✓ Correct

Explanation

QnA Maker can automatically generate Q&A pairs from documentation, and when combined with Form Recognizer for structured document extraction, it provides an efficient pipeline for knowledge base creation from existing materials.

Q15 Hard

You are designing a solution to detect faces in images and verify identity. What privacy consideration must you address?

A Obtain explicit consent and implement data retention policies ✓ Correct
B Use facial recognition without user notification for security
C Store face recognition results in a database for future matching
D Encrypt all facial data but maintain indefinite retention

Explanation

Privacy regulations require explicit user consent for facial recognition and strict data retention policies. Storing sensitive biometric data indefinitely without consent violates responsible AI principles and regulatory requirements.

Q16 Medium

Your solution needs to handle domain-specific terminology that standard language models don't recognize well. What should you implement?

A Implement manual pre-processing of all input text
B Switch to a different language service
C Custom vocabulary and phrase lists in Language Understanding ✓ Correct
D Rely on the default language model only

Explanation

Custom vocabulary and phrase lists in Language Understanding (Conversational LU) allow you to teach the service about domain-specific terms, improving accuracy for industry-specific applications without building a custom language model from scratch.

Q17 Easy

You need to implement a solution that analyzes video content to extract insights. What is the most comprehensive Azure service for this task?

A Azure Media Services for video encoding only
B Azure Computer Vision with frame extraction
C Azure Stream Analytics for real-time video processing
D Azure Video Indexer ✓ Correct

Explanation

Azure Video Indexer provides comprehensive video analysis including scene detection, OCR, speech recognition, object tracking, and custom models—all in a single integrated service specifically designed for video intelligence.

Q18 Medium

Your organization needs to deploy multiple AI models to production with A/B testing capabilities. What approach minimizes risk while gathering performance data?

A Deploy Model A to 100% of users, then deploy Model B after evaluation
B Deploy both models to separate environments without traffic routing
C Deploy the new model to 50% of users immediately for parity testing
D Implement canary deployment sending small traffic percentage to new model ✓ Correct

Explanation

Canary deployment sends a small percentage of traffic to the new model, allowing you to monitor performance and detect issues before full rollout. This minimizes risk while gathering real-world performance data.

Q19 Hard

You are implementing a solution using Azure Cognitive Search. How would you ensure that search results respect user permissions and data classification?

A Use separate search indices for each permission level
B Store all data without security filtering at the search index level
C Manually filter results in the application after search returns
D Implement row-level security using filter expressions in queries ✓ Correct

Explanation

Azure Cognitive Search supports row-level security through filter expressions that enforce permissions at query time, ensuring users only see results they have access to. This is more efficient than post-processing or maintaining separate indices.

Q20 Medium

You need to build a conversational AI solution that maintains context across multiple turns. What architecture element is essential?

A Single-turn question answering without state
B Language Understanding without conversation state
C Pre-computed responses for all possible questions
D Conversation history storage with turn-by-turn context management ✓ Correct

Explanation

Maintaining conversation history and context across turns requires storing previous messages and managing dialogue state. This allows the system to understand pronouns, references, and follow-up questions appropriately.

Q21 Hard

Your solution must process documents in multiple languages where language changes within single documents. How should you handle this?

A Implement Language Understanding with multilingual models
B Detect language per document and process with one language service
C Detect language at paragraph or sentence level and process accordingly ✓ Correct
D Translate entire document to English before processing

Explanation

Detecting language at a granular level (paragraph or sentence) and processing each section with the appropriate language service ensures accurate analysis of mixed-language documents without information loss from translation.

Q22 Hard

You are implementing an AI solution that must maintain consistency in outputs across multiple Azure regions. What is the recommended approach?

A Process all requests sequentially in one region regardless of latency
B Use a single region for all processing to ensure consistency
C Deploy models to multiple regions with Azure Front Door and consistent versioning ✓ Correct
D Allow each region to maintain independent model versions

Explanation

Deploying to multiple regions with Azure Front Door for routing and maintaining consistent model versions ensures both high availability and output consistency. This approach balances performance with reproducibility.

Q23 Medium

You need to implement model retraining to address data drift in your deployed Azure ML model. What should you establish first?

A Automated data drift detection with monitoring and retraining triggers ✓ Correct
B Manual monthly retraining regardless of model performance
C Retraining only after receiving customer complaints
D Continuous retraining on every new data point

Explanation

Automated data drift detection with configured thresholds and automated retraining triggers ensures your model stays accurate as data patterns change. This proactive approach prevents performance degradation.

Q24 Medium

Your organization needs to integrate AI capabilities into an existing legacy application with minimal code changes. Which approach is most practical?

A Completely rewrite the legacy application to use modern AI frameworks
B Abandon the legacy application and build new solution from scratch
C Wrap Azure AI services with an API layer that matches legacy application interfaces ✓ Correct
D Deploy separate AI systems that don't integrate with legacy code

Explanation

Creating an API wrapper layer around Azure AI services that matches legacy application interfaces allows seamless integration without requiring major refactoring. This minimizes disruption while enabling AI capabilities.

Q25 Hard

You are designing an AI solution that processes sensitive healthcare data. Beyond encryption, what additional safeguard is critical?

A Implementation of audit logging and compliance with HIPAA requirements ✓ Correct
B Storing decrypted data in temporary files for processing
C Using public endpoints for accessibility
D Disabling all security features to improve performance

Explanation

Healthcare data requires comprehensive audit logging to track access and modifications, along with compliance to HIPAA requirements. This provides accountability and demonstrates due diligence in protecting sensitive information.

Q26 Medium

You need to implement custom named entity recognition for domain-specific entities not covered by standard models. What is the best approach?

A Use regex patterns and keyword matching for entity extraction
B Manually identify and tag entities without machine learning
C Request Microsoft to add your entities to the standard model
D Train a custom Language Understanding model with labeled examples of your entities ✓ Correct

Explanation

Training a custom Language Understanding model with labeled examples of domain-specific entities allows the system to learn contextual patterns and relationships, providing better accuracy than simple pattern matching or manual tagging.

Q27 Medium

You are designing an Azure AI solution that must classify customer support tickets into categories using text analytics. Which Azure service should you use as the primary component?

A Azure Stream Analytics with pattern matching
B Azure Text Analytics with custom classification ✓ Correct
C Azure Bot Service only
D Azure Cognitive Search with custom scoring profiles

Explanation

Azure Text Analytics provides built-in and custom text classification capabilities specifically designed for categorizing text documents. While other services can support this, Text Analytics is the primary and most direct solution.

Q28 Easy

Your organization needs to extract named entities (persons, locations, organizations) from medical documents. Which Azure service feature should you implement?

A Azure Form Recognizer layout analysis
B Azure Cognitive Search term extraction
C Azure Text Analytics Named Entity Recognition (NER) ✓ Correct
D Azure Language Understanding (LUIS) intent classification

Explanation

Named Entity Recognition (NER) in Azure Text Analytics is specifically designed to identify and extract named entities like persons, organizations, and locations from text documents.

Q29 Medium

You are implementing a solution that must detect faces in images and extract face attributes such as age and emotion. You need to ensure the solution complies with responsible AI principles. Which additional consideration is most critical?

A Use only the highest resolution images to maximize accuracy
B Store all face images in a public blob container for accessibility
C Implement transparency about face attribute accuracy limitations and obtain informed consent for processing ✓ Correct
D Disable age and emotion detection entirely to avoid bias

Explanation

Responsible AI principles require transparency about AI system capabilities and limitations, plus informed consent when processing biometric data. This is more important than technical optimization or removing functionality entirely.

Q30 Medium

You need to implement a solution that recognizes handwritten text from historical documents. Which Azure service is most appropriate?

A Azure Text Analytics language detection
B Azure Form Recognizer with handwritten text recognition model ✓ Correct
C Azure Computer Vision v3.2 OCR
D Azure Cognitive Search with optical character indexing

Explanation

Azure Form Recognizer has specialized models for recognizing handwritten text in documents, including historical and degraded documents. While Computer Vision OCR works, Form Recognizer is more specialized for this use case.

Q31 Medium

Your application requires real-time sentiment analysis of social media posts. The solution must process thousands of posts per minute with minimal latency. What architecture pattern should you implement?

A Batch processing using Azure Data Factory nightly jobs
B Event-driven architecture using Event Hubs with Text Analytics ✓ Correct
C Synchronous REST API calls to Text Analytics for each post
D Store posts in SQL Database and process them weekly

Explanation

Event Hubs enables high-throughput, low-latency processing of streaming data. Batch and synchronous approaches would not meet the real-time requirements with thousands of posts per minute.

Q32 Hard

You are building a knowledge extraction solution from business documents. The solution must identify key phrases, entities, and relationships between entities. Which combination of Azure services best addresses this requirement?

A Computer Vision document analysis with manual relationship mapping
B LUIS for all entity and relationship extraction
C Text Analytics for entities and key phrases, plus a custom entity linking model or knowledge graph ✓ Correct
D Cognitive Search full-text search with aggregations

Explanation

Text Analytics provides entity and key phrase extraction, while relationship extraction requires either custom models or integration with a knowledge graph solution. LUIS is designed for conversational intent, not document analysis.

Q33 Easy

Your organization requires a solution to monitor content moderation across user-generated content. Which Azure service provides built-in content moderation capabilities?

A Azure Text Analytics sentiment analysis
B Azure Cognitive Search content filtering
C Azure Bot Service conversation filtering
D Azure Content Moderator with text, image, and video moderation APIs ✓ Correct

Explanation

Azure Content Moderator is specifically designed for content moderation across text, images, and videos, detecting potentially offensive content. Other services provide analysis but not specialized moderation.

Q34 Medium

You need to implement a multilingual chatbot that understands user intent in 15 different languages. What is the most efficient approach?

A Build custom machine learning models for each language using Azure Machine Learning
B Use LUIS with language-specific models and integrate with Azure Translator Text API for input normalization ✓ Correct
C Create separate LUIS apps for each language and manually switch between them
D Use Azure Bot Service only without any language understanding components

Explanation

LUIS supports multilingual training and intent recognition. Using Translator API for preprocessing can normalize input and improve consistency across languages, making this more efficient than separate apps.

Q35 Medium

Your application must perform optical character recognition (OCR) on business documents with complex layouts, tables, and handwriting. Which Azure service is most suitable?

A Azure Computer Vision Read API v3.2
B Azure Cognitive Search document indexing
C Azure Text Analytics language detection
D Azure Form Recognizer with layout model ✓ Correct

Explanation

Azure Form Recognizer's layout model is specifically designed to handle complex document structures including tables and mixed content. Computer Vision Read API is good for general OCR but less optimized for complex layouts.

Q36 Medium

You are implementing a solution that must detect and redact personally identifiable information (PII) from customer service transcripts. Which approach is most appropriate?

A Manually review all transcripts and apply string replacement
B Store transcripts in SQL Server with encryption and avoid analysis
C Use Text Analytics PII detection and integrate with Azure Data Factory for automated redaction ✓ Correct
D Use Computer Vision to detect text and manually identify PII

Explanation

Text Analytics includes PII detection capabilities that can identify various types of personally identifiable information. Automating this with Data Factory enables scalable, consistent redaction across many documents.

Q37 Medium

Your organization needs to build a custom text classification model that categorizes product reviews as positive, negative, or neutral with high accuracy. Your labeled dataset contains 5,000 reviews. Which approach should you use?

A Azure Text Analytics custom text classification trained on your labeled dataset ✓ Correct
B Use LUIS with custom entities to approximate sentiment categories
C Build a custom Python model using scikit-learn and deploy to Azure Container Instances
D Use Text Analytics built-in sentiment analysis and map results to your categories

Explanation

Azure Text Analytics now offers custom text classification specifically designed for this scenario, where you can train on your own labeled data to create domain-specific classifiers with 5,000 samples being sufficient.

Q38 Medium

You are designing an image analysis solution that must identify objects, read text, and detect faces from real-time video streams. Which service combination is most cost-effective?

A Deploy separate single-service accounts for Computer Vision and Face API
B Implement custom ML models using Azure Machine Learning for cost savings
C Use only Computer Vision for all tasks without Face API
D Use Computer Vision for object detection and text reading; Face API for face detection, sharing an Azure Cognitive Services multi-service account ✓ Correct

Explanation

A multi-service Cognitive Services account provides the best cost efficiency when using multiple services. This single account can handle Computer Vision and Face API operations with consolidated billing.

Q39 Medium

Your solution must integrate speech-to-text conversion with language understanding to build a voice-activated command system. What is the recommended architecture?

A Use Azure Speech Translation service which handles both transcription and intent detection
B Implement custom speech recognition models using Azure Machine Learning
C Azure Speech-to-Text service feeding output to LUIS for intent recognition ✓ Correct
D Use only Azure Bot Service with built-in speech capabilities

Explanation

Azure Speech service handles transcription while LUIS specializes in intent recognition from the transcribed text. This two-stage pipeline is more flexible and accurate than trying to detect intent directly from audio.

Q40 Medium

You need to implement a solution that extracts structured data from invoices with variable layouts and formats. The solution must handle 1,000s of invoices monthly with minimal manual intervention. Which service should you use?

A Azure Computer Vision OCR with manual data mapping
B Azure Cognitive Search with custom analyzers
C Azure Form Recognizer with a trained custom invoice model ✓ Correct
D Azure Data Factory with SQL pattern matching

Explanation

Azure Form Recognizer's custom model capability is specifically designed for extracting structured data from forms and invoices with variable layouts, enabling automated processing at scale.

Q41 Hard

Your organization must comply with data residency requirements where all AI processing must occur in a specific geographic region. Which consideration is most important when selecting Azure AI services?

A Store data in one region but process it in the nearest available region for performance
B Implement your own custom models in Azure Machine Learning to bypass service availability constraints
C Verify that the required cognitive services are available in your target region before designing the solution ✓ Correct
D Use only services that are available in all Azure regions globally

Explanation

Not all Azure Cognitive Services are available in all regions. Before designing your solution, you must verify that the specific services you need are deployed in your required region to meet compliance requirements.

Q42 Medium

You are implementing a solution to analyze customer feedback across multiple channels (email, chat, social media). The solution must identify sentiment, topics, and trends. Which service primarily enables topic analysis?

A Computer Vision image analysis
B Text Analytics with key phrase extraction and custom topic modeling ✓ Correct
C LUIS intent classification
D Azure Cognitive Search term extraction only

Explanation

Text Analytics key phrase extraction helps identify important topics. For more advanced topic modeling, you can combine this with custom machine learning models or unsupervised learning approaches on Azure.

Q43 Medium

Your solution must detect anomalies in time-series data from IoT sensors and trigger alerts when unusual patterns are detected. Which Azure service is most appropriate?

A Azure Stream Analytics with custom query rules only
B Azure Anomaly Detector with multivariate analysis capabilities ✓ Correct
C Azure Cognitive Search with numeric filtering
D Azure Text Analytics with pattern detection

Explanation

Azure Anomaly Detector is specifically designed to detect anomalies in time-series data, including univariate and multivariate patterns. It's more suitable than Stream Analytics rules or Text Analytics for this use case.

Q44 Easy

You need to build a solution that identifies and extracts tables from document images. The documents vary in quality and contain both typed and handwritten content. Which service should you primarily use?

A Azure Computer Vision OCR with post-processing
B Azure Cognitive Search document indexing
C Azure Machine Learning custom vision model
D Azure Form Recognizer with table detection capability ✓ Correct

Explanation

Azure Form Recognizer specifically includes table detection and extraction capabilities, handling both typed and handwritten content in documents. This is more specialized than general OCR approaches.

Q45 Hard

Your organization requires a solution to implement responsible AI practices in a custom computer vision model. Which practices should you implement? (Select the most comprehensive approach)

A Remove all demographic data from training to prevent bias
B Conduct fairness assessments across demographic groups, implement transparency documentation, establish model monitoring, and define acceptable use policies ✓ Correct
C Use only publicly available datasets to ensure fairness
D Achieve the highest possible accuracy by training on the largest available dataset

Explanation

Responsible AI requires a holistic approach including fairness assessment, transparency, monitoring, and governance policies. Simply maximizing accuracy, using public data, or removing demographics are incomplete approaches.

Q46 Medium

You are designing a chatbot that must maintain conversation context across multiple turns and retrieve relevant information from a knowledge base. Which Azure services should you integrate?

A Cognitive Search with direct user queries
B LUIS application without additional services
C Azure Bot Service only with built-in knowledge base
D Azure Bot Framework with LUIS for intent recognition and Cognitive Search for knowledge retrieval ✓ Correct

Explanation

A comprehensive chatbot requires the Bot Framework for orchestration, LUIS for understanding user intent, and Cognitive Search for retrieving relevant information from a knowledge base. Each component serves a specific function.

Q47 Hard

Your solution must process videos to extract frames, analyze each frame for objects and activities, and generate video summaries. What is the most efficient architecture?

A Process entire videos directly with Computer Vision API
B Use Azure Media Services for video processing and Computer Vision for frame analysis with results aggregated for summarization ✓ Correct
C Use Azure Video Analyzer for frame extraction and manual analysis
D Store videos in Azure Storage and use Cognitive Search for indexing

Explanation

Azure Media Services efficiently extracts frames from video at scale, Computer Vision analyzes each frame, and aggregated results enable intelligent summarization. This separates concerns appropriately.

Q48 Easy

You need to implement language translation in your application supporting 50+ language pairs. Which approach minimizes development effort and maintenance?

A Build custom translation models for each language pair using Azure Machine Learning
B Use Azure Translator Text API which natively supports 50+ language pairs and handles encoding automatically ✓ Correct
C Integrate with third-party translation services for unsupported languages
D Implement machine translation models locally for offline functionality

Explanation

Azure Translator Text API natively supports 50+ language pairs with continuous improvements and updates managed by Microsoft. Building custom models would require significant effort for minimal benefit in most cases.

Q49 Hard

Your organization must analyze unstructured documents to extract compliance-relevant information while maintaining audit trails of all processing. Which architecture best supports this requirement?

A Use Cognitive Search without additional logging infrastructure
B Manually review documents and store findings in Excel
C Use Form Recognizer for document analysis with logging to Azure Monitor and storage of processed results in Azure SQL with audit tables ✓ Correct
D Process documents with Text Analytics and store results in blob storage only

Explanation

This architecture provides document analysis through Form Recognizer, maintains audit trails via Azure Monitor logging, and stores structured results in a database that supports audit table tracking. This meets compliance requirements.

Q50 Hard

You are implementing a solution that must handle personal data according to GDPR requirements, including the right to deletion. Which consideration is most critical for your AI solution design?

A Anonymize all data immediately after collection to satisfy GDPR requirements
B Implement data minimization, maintain audit logs, enable data deletion capabilities, and assess privacy impact before deploying AI models on personal data ✓ Correct
C Avoid using personal data in any AI models entirely
D Store personal data only in encrypted containers without processing

Explanation

GDPR compliance requires multiple controls: collecting only necessary data, maintaining audit trails, enabling deletion, and conducting privacy impact assessments. Complete anonymization or avoiding personal data entirely is not always practical or necessary.

Q51 Hard

You are implementing a custom Named Entity Recognition (NER) solution using Azure Machine Learning. You need to train a model to identify custom entities specific to your industry. Which approach should you use to achieve the best accuracy with minimal labeling effort?

A Use Azure Cognitive Search indexers to automatically extract entities without training
B Use Azure Text Analytics pre-built entity recognition and fine-tune with your custom labels
C Train a custom NER model using Azure Machine Learning with transfer learning from a pre-trained BERT model ✓ Correct
D Implement regex patterns to identify all custom entities in your documents

Explanation

Transfer learning with pre-trained BERT models is the most effective approach for custom NER tasks, requiring less labeled data while achieving high accuracy. Azure ML supports this approach through custom training pipelines.

Q52 Medium

You are deploying a real-time language understanding service using Azure Cognitive Services. The service must handle 10,000 requests per minute with guaranteed response times under 100ms. What should you configure?

A Standard pricing tier with auto-scaling enabled
B Premium pricing tier with multiple regional deployments and traffic manager ✓ Correct
C Standard tier with a single deployment in the nearest region
D Free tier with multiple service instances behind a load balancer

Explanation

Premium tier with multi-regional deployment and traffic management ensures the required throughput (10,000 req/min) and low latency (<100ms) through geographic redundancy and higher capacity limits.

Q53 Hard

Your organization needs to process sensitive financial documents using Azure Form Recognizer. The documents contain personally identifiable information (PII) that must not be stored in Azure datacenters. Which configuration best addresses this requirement?

A Use Form Recognizer API with only the analysis results returned, configure data residency settings
B Use Form Recognizer with customer-managed keys (CMK) for encryption at rest
C Process documents on-premises using Form Recognizer Docker containers with local processing ✓ Correct
D Store encrypted documents in Azure Blob Storage with limited access policies

Explanation

Form Recognizer Docker containers enable on-premises processing, ensuring sensitive data never leaves the organization's infrastructure while still leveraging the model's capabilities.

Q54 Medium

You are building a chatbot using Azure Bot Service that integrates with QnA Maker for knowledge management. Users report that the bot frequently provides irrelevant answers. What is the most likely cause and solution?

A The knowledge base needs active learning enabled to improve question-answer matching
B The bot service tier is too basic; upgrade to Premium to improve response quality
C The QnA Maker knowledge base confidence threshold is too low; increase it to filter low-confidence responses ✓ Correct
D The Bot Service is not using LUIS integration for intent recognition

Explanation

A low confidence threshold causes the QnA Maker to return matches that aren't truly relevant. Increasing the threshold filters out low-confidence answers, improving response quality.

Q55 Medium

You are implementing a content moderation solution using Azure Content Moderator for user-generated images. The API must classify images with high precision to avoid false positives. Which configuration should you implement?

A Use automated moderation with maximum sensitivity settings to catch all violations
B Integrate with Computer Vision API only without using Content Moderator
C Use the standard Content Moderator API with default scoring thresholds
D Configure custom review teams and implement human-in-the-loop review for borderline scores with custom threshold adjustments ✓ Correct

Explanation

Human-in-the-loop review with custom threshold configuration balances automation with human judgment, reducing false positives while maintaining moderation effectiveness for borderline cases.

Q56 Easy

You need to implement a solution that translates documents while preserving formatting and layout. Azure Translator Text API alone cannot maintain complex formatting. What is the best approach?

A Use Azure Cognitive Services Speech service for audio translation only
B Export documents to plain text, use Translator API, then manually reformat
C Implement custom Python scripts to parse and translate document elements individually
D Use Document Translation API with batch processing for formatted document translation ✓ Correct

Explanation

Azure's Document Translation API specifically handles formatted documents while preserving layout, structure, and formatting across multiple file formats in batch operations.

Q57 Medium

Your organization is developing a custom vision solution to classify defects in manufacturing quality control. You have 200 labeled images per defect category. What training approach will provide the best results with this limited dataset?

A Train a custom model directly using Custom Vision Service with data augmentation enabled ✓ Correct
B Implement a simple linear regression model to reduce overfitting
C Use transfer learning with a pre-trained ResNet model from Azure Machine Learning Designer
D Collect more images until you have at least 1000 per category before training

Explanation

Custom Vision Service is specifically optimized for small datasets (as few as 50-100 images per category) and includes built-in data augmentation, making it ideal for manufacturing defect classification with limited labeled data.

Q58 Hard

You are designing a semantic search solution using Azure Cognitive Search combined with embeddings. The search index must support semantic ranking of results. Which configuration is required?

A Implement full-text search indexing with BM25 relevance scoring only
B Use Azure OpenAI embeddings integrated with Cognitive Search semantic ranking capability ✓ Correct
C Configure Cognitive Search with custom analyzers and linguistic processors
D Enable semantic search in the Azure Cognitive Search configuration and use text embeddings with vector search

Explanation

Azure Cognitive Search's semantic ranking capability, combined with Azure OpenAI embeddings for vector search, provides advanced semantic understanding and relevance ranking for search results.

Q59 Medium

You need to implement real-time anomaly detection for IoT sensor data using Azure Anomaly Detector. The service must adapt to seasonal patterns in your data. What approach should you use?

A Manually define threshold rules based on historical data patterns
B Implement Anomaly Detector with the `seasonality` parameter set and use the Univariate Anomaly Detection endpoint with appropriate granularity ✓ Correct
C Use Anomaly Detector API with default sensitivity settings
D Build a custom machine learning model to handle seasonality separately

Explanation

Azure Anomaly Detector's `seasonality` parameter explicitly handles seasonal patterns in time-series data, and the Univariate endpoint is optimized for IoT sensor data with configurable granularity.

Q60 Hard

Your organization plans to deploy an AI solution across multiple Azure regions for compliance and latency requirements. You need to implement a strategy for managing and versioning trained models across regions. What is the best practice?

A Use Azure Machine Learning registries for centralized model management and deploy via Azure Container Registry with geo-replication ✓ Correct
B Manually copy trained models to each region's blob storage and update connection strings
C Train separate independent models for each region to ensure compliance isolation
D Store models in Azure Model Registry with region-specific endpoints and implement continuous deployment pipelines using Azure DevOps

Explanation

Azure ML registries provide centralized model versioning and management, while Azure Container Registry's geo-replication ensures models are available and compliant across multiple regions with automatic synchronization.

AI-102 — Designing and Implementing a Microsoft Azure AI Solution Study Guide

About the AI-102 Exam

60 Practice Questions & Answers

Ready to test your knowledge?