Microsoft AI-102 Azure AI Engineer Practice Test

AI-102 is the older Microsoft Azure AI Engineer Associate exam route for developing AI solutions in Azure. Microsoft Learn retirement materials now point learners from the AI-102 training route toward AI-103, the newer Azure AI apps-and-agents route.

This route has been replaced or renamed. Use this page to try 12 transition-oriented sample questions, choose the current equivalent, and avoid starting the wrong Microsoft AI path.

Current equivalent

Why the route changed

Microsoft’s newer AI route emphasizes production AI apps and agents, Microsoft Foundry, generative AI, RAG, multimodal workflows, responsible AI controls, monitoring, and Python-backed implementation. That is a more current fit for candidates building AI apps and agentic systems than a generic older Azure AI Engineer route.

Practice options

• Current status: replaced or retiring route
• IT Mastery coverage for AI-102: replacement guidance page
• Best use right now: try the 12 sample questions, then move from AI-102 preparation to the current AI-103 route
• Recommended next page: AI-103
• Quick review: open the AI-102 cheat sheet to map older Azure AI Engineer concepts to the current AI-103 route.

Sample Exam Questions

Try these 12 original sample questions for Microsoft AI-102. They are designed for self-assessment and to help older AI-102 candidates move toward the current AI-103 Azure AI apps-and-agents path. They are not official exam questions.

Question 1

Topic: replacement fit

A learner searched for AI-102 because their old study plan says “Azure AI Engineer Associate,” but their employer now wants skills in Microsoft Foundry agents and production generative AI apps. What is the best next page to open?

• A. AI-103, because it is the current Azure AI apps-and-agents route.
• B. DP-900, because it covers general Azure data fundamentals.
• C. AZ-140, because it covers virtual desktop administration.
• D. MB-280, because it covers Dynamics 365 customer experience.

Best answer: A

Explanation: Candidates using the older AI-102 route should verify whether AI-103 is now the better target. AI-103 aligns more directly with Azure AI apps, agents, Foundry, and modern generative AI implementation.

What this tests: Choosing the current exam from a legacy exam-code search.

Question 2

Topic: Azure AI service selection

A team needs to extract key fields from scanned invoices and route uncertain results for human review. Which Azure AI capability is most relevant?

• A. Azure Virtual Desktop.
• B. Document intelligence with review workflow controls.
• C. Azure DNS private zones.
• D. Azure Policy exemptions.

Best answer: B

Explanation: Invoice extraction is a document-understanding problem. The strongest option combines extraction with a review process for uncertain or low-confidence fields.

What this tests: Matching an AI workload to the right Azure AI service family.

Question 3

Topic: responsible AI controls

A chatbot sometimes produces unsupported claims from a knowledge base. What should the team add first?

• A. More virtual networks.
• B. A larger VM size.
• C. Grounding, citations, content filtering, and evaluation against expected answers.
• D. A shorter resource group name.

Best answer: C

Explanation: Hallucination and unsupported output are quality and safety issues. Grounding, source citation, content filtering, and evaluation reduce risk better than infrastructure-only changes.

What this tests: Applying responsible AI controls to generated responses.

Question 4

Topic: computer vision workflow

An app must identify damaged products on a conveyor belt using labeled images from the factory. What is the best implementation direction?

• A. Store the images only in Azure Files and manually inspect them.
• B. Use a speech-to-text service because the input is visual.
• C. Replace all cameras with manual reports.
• D. Train or configure a vision model using representative labeled images and validate it against factory examples.

Best answer: D

Explanation: The requirement is visual classification or detection. A representative labeled dataset and validation against real factory conditions are central to a reliable solution.

What this tests: Recognizing image-based AI solution design.

Question 5

Topic: knowledge grounding

A support assistant must answer only from approved internal articles. What design choice best reduces unsupported answers?

• A. Retrieve relevant approved content and ground the model response in those sources.
• B. Allow the model to answer from general internet knowledge.
• C. Disable monitoring so users are not distracted by warnings.
• D. Put every article title in the app name.

Best answer: A

Explanation: Grounded retrieval helps constrain answers to approved content. This is a core pattern in modern Azure AI apps and a bridge from older AI-102 concepts to AI-103-style work.

What this tests: Grounding generated answers in trusted sources.

Question 6

Topic: model evaluation

After launch, user complaints show that an AI extractor misses fields for one document type. What should the team do?

• A. Ignore the issue because the model worked during the demo.
• B. Collect representative failing examples, evaluate the model, and adjust training or configuration.
• C. Increase the subscription spending limit only.
• D. Delete the monitoring dashboard.

Best answer: B

Explanation: AI systems need evaluation and iteration. Real failing examples help improve extraction quality and prevent unsupported assumptions about model performance.

What this tests: Using operational feedback to improve AI quality.

Question 7

Topic: language workloads

A company needs to identify sentiment and key phrases from customer comments in multiple languages. Which solution direction fits best?

• A. Use only Azure Blob lifecycle management.
• B. Build a VM scale set with no AI service.
• C. Use Azure language capabilities that support sentiment and key phrase extraction.
• D. Use Azure Bastion because comments are text.

Best answer: C

Explanation: Sentiment and key phrase extraction are natural-language workloads. Service fit matters more than unrelated infrastructure services.

What this tests: Recognizing language-service use cases.

Question 8

Topic: AI app monitoring

An AI app is deployed to production. Which signal is most useful for detecting quality drift?

• A. The number of resource groups in the subscription.
• B. The color of the app icon.
• C. The names of developers on the team.
• D. Trends in user feedback, grounded-answer quality, failed tasks, and evaluation metrics.

Best answer: D

Explanation: Production AI quality should be monitored through behavior and evaluation signals. Generic subscription inventory does not show whether answers remain useful and safe.

What this tests: Monitoring AI quality after deployment.

Question 9

Topic: security and identity

A service needs to call an Azure AI resource without storing a secret in source code. What is the best approach?

• A. Use managed identity or another supported identity-based access pattern.
• B. Commit the key to the repository so deployment is easier.
• C. Email the key to every developer.
• D. Disable authentication on the resource.

Best answer: A

Explanation: Modern Azure implementation favors identity-based access and secret minimization. Hard-coded keys are a security risk.

What this tests: Secure access patterns for Azure AI resources.

Question 10

Topic: user feedback loop

Users can mark AI answers as helpful or not helpful. What should the product team do with this signal?

• A. Treat every negative rating as proof that AI cannot work.
• B. Use feedback with logs and evaluation data to prioritize improvements.
• C. Delete negative feedback before review.
• D. Replace all validation with team opinions.

Best answer: B

Explanation: Feedback is a useful signal but should be interpreted with other evidence. It can guide prompt, retrieval, safety, and workflow improvements.

What this tests: Combining user feedback with AI evaluation data.

Question 11

Topic: solution boundary

A stakeholder asks the AI app to make final medical decisions without human review. What is the safest response?

• A. Accept the requirement because automation always improves safety.
• B. Remove all disclaimers to make the system more confident.
• C. Identify risk, regulatory, and human-oversight requirements before designing the workflow.
• D. Use a cheaper model and ignore the decision boundary.

Best answer: C

Explanation: High-risk decisions require careful governance, human oversight, and compliance review. AI solution design must respect the decision boundary.

What this tests: Recognizing responsible-use limits.

Question 12

Topic: current study path

A candidate is early in Azure AI and has no implementation experience. What is the best sequence?

• A. Choose a Dynamics 365 exam because it has an AI label.
• B. Skip fundamentals and memorize retired exam names.
• C. Study only database backup terms.
• D. Start with a fundamentals route, then move to AI-103 when implementation and agent work become the target.

Best answer: D

Explanation: Fundamentals can build vocabulary before implementation-heavy AI apps and agents. AI-102 route choice should be checked based on the candidate’s current target.

What this tests: Choosing a realistic Microsoft AI study path.

AI-102 transition map

Use this map to connect the sample questions to the decision pattern Microsoft usually tests for this route.

    flowchart LR
	  S1["Legacy AI-102 route"] --> S2
	  S2["Confirm current AI route"] --> S3
	  S3["Compare AI-103 apps and agents"] --> S4
	  S4["Review Azure AI services"] --> S5
	  S5["Practise responsible implementation"] --> S6
	  S6["Notify me for updates"]

Quick Cheat Sheet

Mini Glossary

• Azure AI Foundry: Microsoft platform surface for building and managing AI apps, models, agents, and evaluations.
• Azure AI Search: Search service often used to retrieve enterprise content for grounded AI responses.
• Grounding: Providing trusted context so an AI answer is based on approved content instead of a bare prompt.
• Replacement route: The current exam or certification path a learner should compare when an older exam is retired or changing.
• Responsible AI: Design practices for safety, transparency, privacy, human oversight, and risk control.

Open Microsoft AI-102 in IT Mastery

Use this page to review sample questions, use the Notify me form for exam-specific updates, and compare related IT Mastery pages.

Official sources

• Microsoft retired courses page
• Microsoft exam retirement list

What to open next

• Use AI-103 if your goal is Azure AI apps and agents.
• Use AI-901 if you need the replacement Azure AI Fundamentals route.

In this section

• Microsoft AI-102 Cheat Sheet: Azure AI Engineer
  Review the legacy Microsoft Azure AI Engineer (AI-102) route, AI service selection, responsible AI, vision, language, search, and the current AI-103 replacement path.