8.5. Practice Questions (71 Questions)
These practice questions mirror the format and difficulty of the actual AI-102 exam. Each question includes a detailed rationale explaining not just the correct answer, but why the distractors are incorrect. Questions are organized by domain to help you identify areas needing additional review.
Domain 1: Plan and Manage an Azure AI Solution (Q1-15)
Q1. You have a website that allows users to upload images. You need to ensure that uploaded images do not contain adult content. The solution must minimize development effort.
Which service should you use?
A. Azure Face in Foundry Tools
B. Azure Vision in Foundry Tools Image Analysis
C. Azure Custom Vision
D. Azure Vision in Foundry Tools Spatial Analysis
Answer: B. Azure Vision in Foundry Tools Image Analysis
Azure Vision Image Analysis includes adult content detection as a built-in feature. Azure Face detects faces, not content types. Custom Vision requires training. Spatial Analysis detects people movement in video.
Q2. Your company plans to develop an app with an AI model using proprietary data. You need to select an Azure platform to ensure the model is connected and operationalized effectively.
Which Azure service should you recommend?
A. Azure Content Understanding in Foundry Tools
B. Microsoft Foundry
C. Azure Speech in Foundry Tools
D. Azure Vision in Foundry Tools
Answer: B. Microsoft Foundry
Microsoft Foundry is the unified platform for deploying models, connecting data, and managing AI resources. Content Understanding focuses on multimodal analysis. Speech and Vision are specific services, not platforms.
Q3. You are building an app that will extract specific information from scanned receipts.
Which service should you use?
A. Azure Application Insights
B. Azure Document Intelligence in Foundry Tools
C. Azure AI Metrics Advisor
D. Azure Language in Foundry Tools
Answer: B. Azure Document Intelligence in Foundry Tools
Document Intelligence extracts structured data from documents using pre-built receipt model. Application Insights monitors applications. Language service processes text, not document extraction.
Q4. You are building an app that will extract insights from video files. You need to identify which service to use. The solution must ensure that you can customize the language model used.
What should you use?
A. Azure Language in Foundry Tools
B. Azure Speech in Foundry Tools
C. Azure AI Video Indexer
D. Azure Vision in Foundry Tools
Answer: C. Azure AI Video Indexer
Video Indexer extracts insights from videos AND supports custom language model customization. Language and Speech don't process video. Vision provides image analysis, not video insights.
Q5. You have an Azure App Services web app named App1. You need to configure App1 to use Microsoft Foundry to authenticate by using Microsoft Entra ID. The solution must minimize administrative effort and use the principle of least privilege.
What should you do?
A. Create a secret and store the secret in Azure Key Vault
B. Enable a managed identity and assign RBAC permissions to Microsoft Foundry
C. Create a certificate and store it in Azure Key Vault
D. Create an API key and hardcode it in App1's configuration
Answer: B. Enable a managed identity and assign RBAC permissions to Microsoft Foundry
Managed identity eliminates credential management with automatic rotation. Key Vault secrets still require rotation. Hardcoded keys are a security risk.
Q6. You have a Microsoft Foundry Service resource. You need to enable diagnostic logging.
What are two prerequisites for diagnostic logging? (Choose two.)
A. A Log Analytics workspace
B. An Azure Cosmos DB for NoSQL account
C. An Azure Storage account
D. An Azure SQL Database
Answer: A, C
Diagnostic logging requires either a Log Analytics workspace (for query/analysis) or a Storage account (for archival). Cosmos DB and SQL Database are not valid destinations.
Q7. You are provisioning an Azure OpenAI resource. You need to ensure the resource is only available to applications in your Azure subscription.
Which network security setting should you configure?
A. All networks
B. Selected networks
C. Disabled, and allow a private endpoint connection
D. Disabled, and configure a firewall rule
Answer: C. Disabled, and allow a private endpoint connection
Private endpoint ensures only Azure subscription applications can access. All networks allows public access. Selected networks allows external access if configured.
Q8. You implement a Content Safety solution. Users report that some images fail to display after being uploaded. You need to ensure proper content flags.
What should you do?
A. Adjust the content categories
B. Adjust the severity level
C. Enable additional content filters
D. Disable Content Safety for PNG files
Answer: B. Adjust the severity level
If legitimate images are blocked, the severity threshold is too low. Adjusting severity balances safety with usability. Changing categories doesn't address false positives.
Q9. You discover attempts to exploit your copilot through a User Prompt Injection Attack (UPIA). You need to increase security.
Which Microsoft Foundry Service should you use?
A. Content Safety
B. Azure Monitor
C. Azure Key Vault
D. Azure Translator
Answer: A. Content Safety
Content Safety includes jailbreak risk detection that recognizes attack patterns. Other services don't address jailbreak prevention.
Q10. You're developing an AI-powered e-learning platform. Concerns exist about generating inappropriate educational materials. You need to ensure content complies with standards.
What should you implement?
A. Enable Azure Translator
B. Implement Microsoft Foundry Content Safety
C. Train a custom model
D. Enable Azure Text Analytics
Answer: B. Implement Microsoft Foundry Content Safety
Content Safety analyzes inputs/outputs to detect and block harmful content. Translator handles translation. Custom models improve relevance but don't ensure safety.
Q11. You're deploying a generative AI model for customer support. You need to ensure the model generates safe outputs.
Which three actions should you take? (Choose three.)
A. Conduct red team exercises
B. Document the model's decision-making logic
C. Enable users to provide feedback on every response
D. Integrate Microsoft Foundry Content Safety APIs
E. Train the model exclusively on synthetic data
Answer: A, B, D
Red team exercises identify vulnerabilities. Documentation supports transparency. Content Safety filters harmful content. Feedback on every response overwhelms users.
Q12. You are creating an assistant using a generative model. You plan to use the system message.
Which two capabilities does the system message offer? (Choose two.)
A. Defines data sources that should not be included
B. Defines what the model should and should not do
C. Helps define the assistant's personality
D. Specifies the model's maximum token output
E. Sets the temperature parameter
Answer: B, C
System message defines personality and behavior rules. Token limits and temperature are set in API parameters, not system message.
Q13. You plan to deploy a generative AI solution using GPT-4. You need to ensure GPT-4 is available for inferencing via an endpoint.
Which three actions should you perform? (Choose three.)
A. Create a new Azure subscription
B. Deploy a GPT-4 model
C. Provision a Microsoft Foundry resource
D. Select GPT-4 from the catalog
E. Set up a virtual machine
Answer: B, C, D
Provisioning the resource, selecting the model, and deploying it are required. New subscription isn't needed. VMs aren't required—Foundry uses managed endpoints.
Q14. You are building a GPT-based chat application. You plan to test using Microsoft best practices.
Which three prompt engineering strategies should you consider? (Choose three.)
A. Be Descriptive
B. Be Minimalistic
C. Be Simple
D. Be Specific
E. Order Matters
Answer: A, D, E
Microsoft best practices: Be Specific, Be Descriptive (use analogies), and Order Matters. Be minimalistic and simple don't produce optimal results.
Q15. Your Azure OpenAI model is set to auto-update. You need consistent behavior during updates and to test new versions before deployment.
Which two actions should you take? (Choose two.)
A. Disable automatic updates
B. Enable updates to the default version
C. Increase deployment resources
D. Select a specific model version
E. Test new versions separately
Answer: D, E
Selecting a specific version prevents unexpected updates. Testing separately validates before production.
Domain 2: Implement Generative AI Solutions (Q16-27)
Q16. You are building a web app to generate images using DALL-E 3. You need HTTP requests to successfully generate images.
Which three HTTP header properties should you include? (Choose three.)
A. The API version
B. The name of the Azure OpenAI resource
C. The name of the DALL-E 3 deployment
D. The prompt text
E. The image size specification
Answer: A, B, C
Resource name, deployment name, and API version are required headers. Prompt is a body property. Image size is optional body.
Q17. You are building a web app using DALL-E 3. Which HTTP body property should you include?
A. The API version
B. The deployment name
C. The prompt
D. The resource name
Answer: C. The prompt
Prompt is the only required body property. API version, deployment name, and resource name are headers.
Q18. You're using Azure OpenAI for document summarization. You need summaries to meet organizational requirements.
What action should you take?
A. Enable diagnostic logging
B. Increase the token limit
C. Refine prompts to specify key details
D. Switch to a higher-cost model
Answer: C. Refine prompts to specify key details
Prompt engineering directly improves output quality. Logging monitors but doesn't improve. Higher token limits allow longer responses but not better quality.
Q19. You're deploying Azure OpenAI to process customer feedback. You need to optimize for performance and ensure responses align with your organization's tone.
Which three actions should you perform? (Choose three.)
A. Deploy the model on edge devices
B. Enable multilingual support
C. Fine-tune the model with customer feedback data
D. Set up monitoring to track response accuracy
E. Use prompt engineering to refine output
Answer: C, D, E
Fine-tuning customizes tone/style. Monitoring tracks performance. Prompt engineering refines output.
Q20. Your text-generation model produces responses that vary in tone. You need to reduce randomness.
Which parameter should you configure?
A. max_tokens
B. role
C. stop
D. temperature
Answer: D. temperature
Temperature controls randomness—lower values produce more consistent outputs. max_tokens limits length. role defines message author.
Q21. You need to implement a RAG solution that grounds Azure OpenAI responses in your corporate data. Which two components are essential? (Choose two.)
A. Azure AI Search
B. Azure AI Video Indexer
C. Azure OpenAI
D. Custom Vision
Answer: A, C
RAG requires Azure AI Search for retrieval and Azure OpenAI for generation. Video Indexer and Custom Vision are not part of the RAG pattern.
Q22. You're configuring Azure OpenAI to return only valid JSON. What should you set?
A. max_tokens to a specific value
B. response_format to {"type": "json_object"}
C. temperature to 0
D. top_p to 0.1
Answer: B. response_format to {"type": "json_object"}
JSON mode ensures the model returns valid JSON. Temperature and top_p control randomness, not format.
Q23. You're building a prompt flow in Azure AI Foundry. You need to evaluate the quality of your flow's outputs.
What should you do?
A. Configure diagnostic logging
B. Create evaluation datasets and run evaluation flows
C. Deploy the flow to production
D. Enable tracing only
Answer: B. Create evaluation datasets and run evaluation flows
Azure AI Foundry provides evaluation capabilities with built-in metrics. Logging and tracing monitor but don't evaluate quality.
Q24. You need to create vector embeddings for semantic search. Which Azure OpenAI model should you use?
A. DALL-E 3
B. GPT-4o
C. text-embedding-ada-002
D. Whisper
Answer: C. text-embedding-ada-002
Embedding models like text-embedding-ada-002 create vector representations. GPT models generate text, DALL-E generates images, Whisper transcribes audio.
Q25. You're implementing Azure OpenAI "On Your Data" feature. What does this enable?
A. Fine-tuning the model with your data
B. Grounding responses in your indexed data
C. Storing conversation history
D. Training a new model from scratch
Answer: B. Grounding responses in your indexed data
"On Your Data" enables RAG by connecting Azure OpenAI to your Azure AI Search index for grounded responses.
Q26. You are using Azure Translator. You need to extend capabilities.
Which three features are available? (Choose three.)
A. Detect language
B. Dictionary lookup
C. Entity extraction
D. Sentiment analysis
E. Transliterate
Answer: A, B, E
Translator includes: Transliterate, Detect, Dictionary lookup. Entity extraction and sentiment are Language service features.
Q27. You are building an Azure Translator custom model. You need a BLEU score indicating high quality.
What is the minimum score range required?
A. 0 to 19
B. 20 to 39
C. 40 to 59
D. 60 to 100
Answer: C. 40 to 59
BLEU scores of 40-59 indicate high quality translation.
Domain 3: Implement an Agentic Solution (Q28-38)
Q28. You're deploying an agent to perform actions like scheduling meetings. You need to integrate with tools that enable programmatic actions.
What should you do?
A. Add custom functions
B. Configure static templates
C. Use default model capabilities
D. Use a pre-trained chatbot framework
Answer: A. Add custom functions
Custom functions enable agents to execute specific actions programmatically.
Q29. You need to select a framework supporting AI orchestration and multi-agent workflows.
What should you recommend?
A. Azure Bot Framework
B. Azure Machine Learning Studio
C. Cognitive Services API
D. Semantic Kernel framework
Answer: D. Semantic Kernel framework
Semantic Kernel supports generative AI orchestration and multi-agent workflows.
Q30. You're building an Azure OpenAI Assistant. You need to enable the assistant to execute Python code.
Which tool should you configure?
A. code_interpreter
B. file_search
C. function
D. retrieval
Answer: A. code_interpreter
The code_interpreter tool allows assistants to execute Python code in a sandboxed environment.
Q31. Your Azure OpenAI Assistant run returns a status of "requires_action". What should you do next?
A. Create a new thread
B. Delete the assistant
C. Execute the requested tool calls and submit outputs
D. Wait for the status to change automatically
Answer: C. Execute the requested tool calls and submit outputs
When status is "requires_action", you must execute the tool calls and use submit_tool_outputs to continue.
Q32. You're implementing a multi-agent solution where different agents handle different tasks. Which approach should you use?
A. Deploy all functionality in a single agent
B. Use AutoGen or Semantic Kernel for agent orchestration
C. Create separate Azure OpenAI resources for each agent
D. Implement agents without coordination
Answer: B. Use AutoGen or Semantic Kernel for agent orchestration
AutoGen and Semantic Kernel provide frameworks for orchestrating multiple agents working together.
Q33. In Autogen, you want an agent to execute code without any human approval. What parameter should you set?
A. code_execution_config={"auto_approve": True}
B. human_input_mode="NEVER"
C. autonomous=True
D. require_approval=False
Answer: B. human_input_mode="NEVER"
The human_input_mode parameter controls when human input is required. "NEVER" enables fully autonomous execution without any human approval steps.
Q34. You're using Semantic Kernel and need to add weather lookup capability to your agent. What should you create?
A. A tool with @tool decorator
B. A plugin with @kernel_function decorator
C. A function with @function_calling decorator
D. A skill with @sk_function decorator
Answer: B. A plugin with @kernel_function decorator
Semantic Kernel uses plugins with the @kernel_function decorator to extend agent capabilities. Tools are Azure OpenAI Assistants terminology, not Semantic Kernel.
Q35. In an Autogen GroupChat, you want the LLM to decide which agent should speak next based on conversation context. What speaker_selection_method should you use?
A. "round_robin"
B. "random"
C. "auto"
D. "sequential"
Answer: C. "auto"
The "auto" method uses the LLM to intelligently select the next speaker based on conversation context. Round_robin cycles through agents in fixed order regardless of context.
Q36. You need to build a multi-agent system where a planner breaks down tasks, a coder implements them, and a reviewer checks the code. Which orchestration pattern is this?
A. Sequential
B. Hierarchical
C. Collaborative
D. Competitive
Answer: B. Hierarchical
Hierarchical orchestration has a manager (planner) delegating to specialized workers (coder, reviewer). Sequential would be fixed order without task-based delegation.
Q37. You're building an enterprise AI application that needs to integrate with existing Azure services and requires governance controls. Which agent framework should you use?
A. Autogen
B. LangChain
C. Semantic Kernel
D. Hugging Face Transformers
Answer: C. Semantic Kernel
Semantic Kernel is Microsoft's framework with native Azure integration and enterprise features. Autogen excels at multi-agent but has less enterprise integration.
Q38. You're testing a Semantic Kernel agent and need to verify it calls the correct plugin for a weather query without making actual API calls. What testing approach should you use?
A. Integration test with live LLM
B. Unit test with mocked kernel services
C. Load test with multiple queries
D. End-to-end test with production data
Answer: B. Unit test with mocked kernel services
Mocking kernel services allows testing plugin invocation logic without actual API calls or costs. Integration tests would make real API calls.
Domain 4: Implement Vision Solutions (Q39-51)
Q39. You need to identify object locations in an image. Which feature should you use?
A. Detect brands
B. Image tagging
C. Object detection
D. Recognize domain-specific content
Answer: C. Object detection
Object detection returns bounding box coordinates. Image tagging returns labels only without locations.
Q40. You are building an app to extract text from scanned handwritten images. Which feature should you use?
A. Azure AI Custom Vision
B. Image classification
C. Object detection
D. Optical character recognition (OCR)
Answer: D. OCR
OCR extracts text from images, including handwritten text.
Q41. You need to analyze and detect animals in images. Which project type should you use?
A. Image classification
B. Object detection
C. Semantic segmentation
D. Image captioning
Answer: B. Object detection
Object detection returns both labels AND coordinates for animals.
Q42. You are creating a model to detect and locate animals in wildlife images. What should you use?
A. Build a CNN model using Azure Machine Learning
B. Train a classification model using Azure Custom Vision
C. Use Azure Custom Vision to train an object detection model
D. Use the prebuilt image tagging model
Answer: C. Use Azure Custom Vision to train an object detection model
Custom Vision object detection identifies AND locates animals.
Q43. You're developing a custom vision model. You need to ensure accuracy.
What action should you take?
A. Deploy the model without testing
B. Evaluate precision and recall metrics
C. Retrain with a new dataset
D. Source new images for testing
Answer: B. Evaluate precision and recall metrics
Precision and recall metrics assess model performance.
Q44. You're developing a model to classify food items with tags 'vegetable-fruit', 'dessert', 'soup'. You need to optimize training.
Which two actions should you perform? (Choose two.)
A. Select the 'Food' domain
B. Select the 'General' domain
C. Set a high probability threshold during training
D. Use consistent tags to label images
Answer: A, D
Food domain optimizes for food classification. Consistent tags ensure accurate training.
Q45. You're using Video Indexer API to analyze Teams recordings. You need to search for competing company mentions.
Which content model should you use?
A. Custom brands
B. Custom Language model
C. Slate detection
D. Topics extraction
Answer: A. Custom brands
Custom Brands model detects brands from speech and visuals.
Q46. You're training a custom Language model in Video Indexer based on word combination probability.
Which three practices should be followed? (Choose three.)
A. Include at least 500,000 sentences
B. Include multiple examples of spoken sentences
C. Include special characters such as ~, #, @
D. Provide multiple adaptation options
E. Put only one sentence per line
Answer: B, D, E
Multiple examples, adaptation options, and one sentence per line are best practices. Special characters are discarded.
Q47. Your Speech Service app has high WER with many substitution errors. What should you add to training data?
A. Background noise samples
B. Custom product and people names
C. Overlapping speaker samples
D. Silence samples
Answer: B. Custom product and people names
Substitution errors indicate need for custom vocabulary. Background noise causes insertion errors.
Q48. You're building a CLU model. You need the metric measuring TP / (TP + FP).
Which metric should you use?
A. BLEU
B. F1 score
C. Precision
D. Recall
Answer: C. Precision
Precision = TP / (TP + FP). Recall = TP / (TP + FN).
Q49. You execute an API call with synonyms including "#diagnostic" and receive an error.
What should you do?
A. Add more synonyms
B. Modify the order of synonyms
C. Remove any special characters
D. Remove duplicate synonyms
Answer: C. Remove any special characters
Special characters (#, @, %) are not allowed in synonyms.
Q50. You're building a chatbot for FAQs. Identify suitable scenarios for question answering service.
Which three scenarios? (Choose three.)
A. Bot conversation with dynamic information
B. Bot conversation with static information
C. Dynamic information in knowledge base
D. Static information in knowledge base
E. Same answer to a request
Answer: B, D, E
Question answering works with static information and consistent answers. Dynamic information is NOT suitable.
Q51. You're building a chatbot using question answering. Identify operational cost factors.
Which two parameters influence costs? (Choose two.)
A. Number of metadata tags
B. Number of knowledge base editors
C. Required throughput
D. Size and number of knowledge bases
Answer: C, D
Throughput and knowledge base size/count affect pricing.
Domain 5: Implement NLP Solutions (Q52-61)
Q52. You're importing FAQ documents. Which types of data will be extracted?
A. Formatted text, URLs, and lists only
B. Formatted text, URLs, images, and lists
C. Plain text only
D. All document elements including embedded images
Answer: A. Formatted text, URLs, and lists only
FAQ import extracts text, URLs, and lists. Images within documents are NOT extracted.
Q53. You're using "Using your data" feature to ground GPT with company data.
Which four file types can you use? (Choose four.)
A. HTML
B. MD
C. PDF
D. TXT
E. XML
F. ZIP
Answer: A, B, C, D
Supports: TXT, MD, HTML, PDF, Word, PowerPoint. XML and ZIP are NOT supported.
Q54. You're building a knowledge mining solution. You need to store enriched output for downstream apps.
What should you create?
A. A data lake
B. A knowledge store
C. A search index
D. A skillset
Answer: B. A knowledge store
Knowledge store persists enriched data for downstream apps.
Q55. Your company collects business cards. You need to extract contact details.
What should you use?
A. Train a custom model
B. Use Azure Vision OCR
C. Use the Business Card model
D. Use the ID Document model
Answer: C. Use the Business Card model
Business Card model is specifically designed for business cards.
Q56. You're processing scanned vendor documents. You need to extract vendor details, totals, and line items.
What should you use?
A. Train a custom model
B. Use Azure Vision OCR
C. Use the Business Card model
D. Use the Invoice model
Answer: D. Use the Invoice model
Invoice model is optimized for extracting structured invoice data.
Q57. Your Document Intelligence app fails on some PDFs (up to 2 MB, 10 pages). What is the cause?
A. File size exceeds limits
B. Page count exceeds limits
C. Some files are password protected
D. S0 tier is insufficient
Answer: C. Some files are password protected
Password-protected files cannot be processed regardless of size.
Q58. You're processing invoices from various vendors. You need to extract key fields.
Which two actions should you perform? (Choose two.)
A. Create a custom model
B. Install on a local server
C. Provision a Document Intelligence resource
D. Use the prebuilt invoice model
Answer: C, D
Provisioning the resource and using prebuilt model are required. Custom model is unnecessary.
Q59. You need to extract information from documents, images, videos, and audio with minimum effort.
What should you use?
A. Azure Content Understanding
B. Azure Document Intelligence
C. Azure Language
D. Azure Machine Learning
Answer: A. Azure Content Understanding
Content Understanding handles documents, images, videos, and audio through a unified API.
Q60. You're processing scanned invoices for financial recordkeeping.
Which three actions should you take? (Choose three.)
A. Ensure input files meet specifications
B. Select the Business Card model
C. Select the Invoice model
D. Select the Read model
E. Use text-based PDF files
Answer: A, C, E
Meeting specifications, using Invoice model, and text-based PDFs optimize extraction.
Q61. You have image files with tables. You need to use the layout model without training.
Which three methods can you use? (Choose three.)
A. A client library SDK
B. Document Intelligence Studio
C. REST API
D. A prebuilt model
E. Custom model training
Answer: A, B, C
SDK, REST API, and Studio can invoke the prebuilt layout model without training.
Domain 6: Implement Knowledge Mining and Document Intelligence (Q62-71)
Q62. You're developing an AI agent for document retrieval. You need to interact with external data sources.
What should you use?
A. Azure Bot Framework
B. Azure AI Search
C. Azure Document Intelligence
D. Azure Machine Learning
Answer: B. Azure AI Search
Azure AI Search provides indexing and querying for external data sources.
Q63. Your Azure AI Search skillset enriches documents. You need to persist data for analytics.
What should you configure?
A. A knowledge store with table projections
B. A secondary search index
C. An Azure Cosmos DB connection
D. Azure Blob storage directly
Answer: A. A knowledge store with table projections
Knowledge store with table projections persists enriched data for analytics.
Q64. You're implementing Azure AI Search. You need to enable AI enrichment during indexing.
What should you create?
A. A custom analyzer
B. A data source
C. A scoring profile
D. A skillset
Answer: D. A skillset
Skillsets define AI enrichment operations during indexing.
Q65. You need to implement semantic search for natural language queries.
What should you configure?
A. A custom analyzer
B. A semantic configuration
C. A vector index
D. Full-text search only
Answer: B. A semantic configuration
Semantic configuration enables ML-based ranking for natural language queries.
Q66. Your Document Intelligence solution needs to support an additional contract format with minimum effort.
What should you do?
A. Add the format to existing training set and retrain
B. Create a new custom model
C. Lower the confidence threshold
D. Use OCR instead
Answer: A. Add the format to existing training set and retrain
Adding to existing training and retraining is efficient. New model duplicates effort.
Q67. You're building a RAG solution.
Which two services should you use together? (Choose two.)
A. Azure AI Search
B. Azure AI Video Indexer
C. Azure OpenAI
D. Azure Translator
Answer: A, C
RAG combines Azure AI Search (retrieval) with Azure OpenAI (generation).
Q68. You need to create vector embeddings for semantic search.
Which Azure OpenAI model should you use?
A. DALL-E 3
B. GPT-4o
C. text-embedding-ada-002
D. Whisper
Answer: C. text-embedding-ada-002
Embedding models create vector representations for semantic search.
Q69. You need to extract data from W-2 forms.
Which model should you use?
A. Business Card model
B. Custom model
C. Invoice model
D. W-2 model
Answer: D. W-2 model
Document Intelligence includes a prebuilt W-2 model for tax forms.
Q70. You're configuring an indexer to process PDFs from Blob Storage. You need to extract text and tables.
Which skill should you include?
A. Entity Recognition skill
B. Language Detection skill
C. OCR skill
D. Sentiment skill
Answer: C. OCR skill
OCR skill extracts text from images and scanned PDFs.
Q71. Your composed document intelligence model fails on a document.
What is the most likely cause?
A. Composed model exceeds maximum component limit
B. Document doesn't match any component model
C. Document is password protected
D. Document size exceeds limits
Answer: C. Document is password protected
Password-protected documents cannot be processed regardless of model type.
