Microsoft · AI-102
Microsoft Certified: Azure AI Engineer Associate (AI-102) Practice Tests
Validates expertise in designing and implementing Azure AI solutions using Azure AI services, Azure AI Search, and Azure OpenAI. Covers computer vision, natural language processing, knowledge mining, generative AI, and agentic solutions.
Exam Details
Practice Questions
1,392
≈ 27 practice exams
Duration
100 minutes
Passing Score
700/1000
Difficulty
AssociateLast Updated
Jan 2026
Topics Covered
Azure AI ServicesComputer VisionNatural Language ProcessingKnowledge MiningGenerative AIAzure OpenAIAzure AI SearchResponsible AI
Microsoft Certified: Azure AI Engineer Associate (AI-102) Practice Exam Preparation
Use this AI-102 practice exam to prepare for Microsoft Certified: Azure AI Engineer Associate (AI-102) with realistic questions, detailed explanations, and focused study modes. The practice bank includes 1,392 questions for Microsoft AI-102, so you can review the exam steadily instead of relying on one long cram session.
As you practice, pay extra attention to recurring topics such as Azure AI Services, Computer Vision, Natural Language Processing, Knowledge Mining, and Generative AI. Start with short sessions to identify weak areas, then move into timed quizzes once your accuracy is consistent.
The explanations are especially useful when you want to connect exam wording to the responsibilities and scenarios described in the official certification guidance. Use the free preview first, then unlock the full question bank when you are ready to build a complete study routine.
Exam Domain Breakdown
Plan and manage an Azure AI solution23%
Implement generative AI solutions18%
Implement an agentic solution8%
Implement computer vision solutions13%
Implement natural language processing solutions18%
Implement knowledge mining and information extraction solutions18%
Exam Overview
The Microsoft Certified: Azure AI Engineer Associate certification, validated through exam AI-102 (Designing and Implementing a Microsoft Azure AI Solution), demonstrates expertise in building, managing, and deploying production-grade AI solutions on the Azure platform. The exam covers the full breadth of Azure AI capabilities including Azure AI Services, Azure AI Search, Azure OpenAI in Foundry Models, Azure AI Vision, Azure AI Language, Azure AI Speech, Azure AI Document Intelligence, and Azure Content Understanding. Candidates are assessed on their ability to architect secure, end-to-end AI solutions that span computer vision, natural language processing, knowledge mining, generative AI, and autonomous agentic workflows.
Last updated on December 23, 2025, the exam reflects Microsoft's shift toward the Microsoft Foundry platform and its emphasis on generative AI and agentic solutions. It aligns with the real-world responsibilities of an Azure AI engineer, including requirements definition, development, deployment, integration, performance tuning, and monitoring. The certification also requires demonstrating the ability to apply Responsible AI principles, configure content safety filters, and design governance frameworks—reflecting increasing enterprise demand for trustworthy AI systems.
Who Should Take This Exam
This certification is designed for software engineers and developers who build AI-integrated applications and want to formalize their expertise on the Azure platform. Ideal candidates work in roles such as AI Engineer, Machine Learning Engineer, Cloud Developer, Solutions Architect, or Data Engineer with a focus on AI. Candidates should have hands-on development experience using Python or C# and be comfortable consuming REST APIs and SDKs across the Azure AI portfolio.
The certification is most relevant to professionals who collaborate with data scientists, data engineers, IoT specialists, and infrastructure teams to integrate AI capabilities into broader enterprise solutions. It suits those with at least one to two years of experience building cloud-based applications on Azure and some familiarity with AI/ML concepts, though it does not require deep data science or model-training expertise.
Prerequisites
There are no mandatory prerequisites to sit for the exam, but Microsoft recommends foundational Azure knowledge equivalent to the AZ-900 (Azure Fundamentals) or AI-900 (Azure AI Fundamentals) certifications before attempting AI-102. Candidates should be proficient in Python or C# and comfortable with REST API consumption and SDK usage in at least one of those languages.
Practical experience is strongly recommended: candidates should have hands-on familiarity with Azure AI Services (including Vision, Language, Speech, and Document Intelligence), Azure AI Search, and Azure OpenAI. Understanding of cloud architecture concepts—such as resource provisioning, identity and access management, containerization, and CI/CD pipelines—will be essential for the planning and management domain. Exposure to prompt engineering, retrieval-augmented generation (RAG) patterns, and responsible AI principles is increasingly important given the exam's updated 2025 objectives.
Exam Format
Exam AI-102 is a proctored assessment delivered through Pearson VUE, available in both online proctored and in-person testing center formats. Candidates are given 100 minutes to complete the exam. The question count typically ranges from 40 to 60 items and may include multiple-choice, multiple-select, drag-and-drop, case studies, and interactive lab-style components. Microsoft does not publish a fixed question count, as it varies per exam form.
The passing score is 700 on a scale of 100–1000. Scoring is scaled and not a simple percentage of correct answers. The exam is available in English, Japanese, Chinese (Simplified and Traditional), Korean, German, French, Spanish, Portuguese (Brazil), and Italian. Candidates who test in a non-English language can request 30 additional minutes. If a candidate fails, they may retake the exam after 24 hours; subsequent retakes require a 14-day waiting period, and no more than five attempts are allowed within a 12-month period. The resulting certification is valid for one year and can be renewed at no cost via an online assessment on Microsoft Learn.
Skills Measured
- 1.Plan and manage an Azure AI solution (20–25%): Selecting appropriate Microsoft Foundry services for generative AI, vision, NLP, speech, and information extraction scenarios; provisioning and deploying Azure AI resources and models; integrating services into CI/CD pipelines; container deployment; monitoring costs and performance; managing authentication and account keys; implementing responsible AI governance including content filters, blocklists, prompt shields, and harm detection.
- 2.Implement generative AI solutions (15–20%): Building solutions with Microsoft Foundry including deploying hubs, projects, and models; implementing prompt flow and retrieval-augmented generation (RAG); evaluating models and flows; integrating with the Microsoft Foundry SDK; using Azure OpenAI in Foundry Models to generate text, code, and images (including DALL-E and large multimodal models); applying prompt engineering techniques; fine-tuning models; configuring model monitoring, tracing, and feedback; and orchestrating multiple generative AI models.
- 3.Implement an agentic solution (5–10%): Understanding agent roles and use cases; configuring resources and creating agents with the Microsoft Foundry Agent Service; implementing complex agents and multi-agent orchestration workflows using Microsoft Agent Framework; enabling autonomous capabilities; and testing, optimizing, and deploying agents.
- 4.Implement computer vision solutions (10–15%): Analyzing images with Azure Vision in Foundry Tools including object detection, tagging, and text extraction (OCR and handwriting); building and consuming custom image classification and object detection models; evaluating and publishing custom vision models; and analyzing videos with Azure AI Video Indexer and Spatial Analysis for detecting people presence and movement.
- 5.Implement natural language processing solutions (15–20%): Extracting key phrases, entities, sentiment, language, and PII from text; translating text and documents with Azure Translator in Foundry Tools; implementing text-to-speech and speech-to-text with Azure Speech in Foundry Tools including SSML, custom speech, intent recognition, and speech translation; and building custom language understanding models and question answering (QnA) solutions including multi-turn conversations and multi-language support.
- 6.Implement knowledge mining and information extraction solutions (15–20%): Provisioning Azure AI Search resources, defining indexes, skillsets, and indexers; implementing custom skills; querying with full-text, semantic, and vector search; managing Knowledge Store projections; using Azure Document Intelligence in Foundry Tools with prebuilt and custom models; training and publishing custom document intelligence models; and using Azure Content Understanding in Foundry Tools to extract text, entities, tables, images, and insights from documents, images, video, and audio.
Study Tips
- Work through the official Microsoft Learn learning path for AI-102, which is structured by exam domain and includes hands-on sandbox exercises. Access it via the 'Two ways to prepare' section on the official certification page.
- Use the free official Practice Assessment (Assessment ID 61) on Microsoft Learn before scheduling your exam. It mirrors the wording, style, and difficulty of real questions and identifies specific knowledge gaps by domain.
- Provision a free-tier or pay-as-you-go Azure subscription and build small end-to-end projects covering each domain: deploy an Azure OpenAI resource and build a RAG pipeline, create a custom vision model, build a QnA solution, and index documents with Azure AI Search. Hands-on lab time is essential.
- Study the official AI-102 Study Guide at aka.ms/ai102-StudyGuide carefully. The December 23, 2025 version reflects current objectives. Pay special attention to the 'Implement an agentic solution' domain (5–10%), which is new and covers Microsoft Foundry Agent Service and multi-agent orchestration with Microsoft Agent Framework.
- Explore the Azure AI Foundry portal (ai.azure.com) hands-on. The exam heavily tests Microsoft Foundry workflows including hub/project deployment, prompt flow, model evaluation, and tracing—capabilities that are best learned by actually using the portal rather than reading about them.
- Review responsible AI documentation including Azure AI Content Safety, prompt shields, content filters, and blocklists. The 'Plan and manage' domain (20–25%) includes a responsible AI governance section that is frequently tested.
- Use the Exam Sandbox at https://aka.ms/examdemo to familiarize yourself with interactive question types including drag-and-drop and case study formats before exam day, so you don't lose time getting oriented during the real exam.
Career Benefits
Earning the AI-102 certification positions professionals for high-demand roles including Azure AI Engineer, Machine Learning Engineer, Cloud AI Developer, AI Solutions Architect, and Cognitive Services Developer. According to ZipRecruiter, the average annual salary for Azure AI Engineers in the United States was approximately $158,110 as of early 2026, with senior roles at top technology companies exceeding $200,000. The certification is particularly valuable as enterprises scale generative AI adoption—engineers who can implement production-ready RAG pipelines, deploy Azure OpenAI solutions, and build agentic workflows are commanding significant salary premiums.
The AI-102 is the primary associate-level AI engineering credential in the Microsoft ecosystem, sitting above AI-900 (Azure AI Fundamentals) and complementing DP-100 (Azure Data Scientist Associate) for teams building end-to-end ML pipelines. Industry surveys indicate that 26% of certified professionals report job promotions and 35% report salary increases following certification. The exam's 2025 update to include agentic AI and Microsoft Foundry aligns the credential with current enterprise priorities, making it more relevant than its predecessors to organizations deploying Azure OpenAI at scale.
Sample Questions
5 sample questions with answers and explanations. The full bank has 1,392 questions, enough for 27 full-length practice exams.
Preview — answers shown1. Contoso Bank is building a RAG-powered internal assistant that answers employee questions using their policy documents indexed in Azure AI Search. During testing, they find that the assistant occasionally answers questions using its general knowledge rather than strictly from the indexed policy documents, leading to potentially incorrect guidance. They need to ensure the assistant only responds based on retrieved policy content. Which configuration change should they make to the Azure OpenAI on your data integration? (Select two!)
Multiple correct answersAEnable the 'Limit responses to your data' setting to prevent the model from using parametric knowledge outside retrieved documents
BSet the RAG strictness parameter to 5 so that only highly relevant retrieved documents are passed to the model as grounding context
CReduce the number of retrieved documents from 10 to 3 to minimize the chance of irrelevant context being included in the prompt
DSwitch the deployment from Standard to Provisioned throughput to reduce latency and improve retrieval accuracy
EEnable groundedness detection as a post-generation filter to flag responses that are not supported by the retrieved source documents
Explanation
Enabling 'Limit responses to your data' instructs the model to only answer from the retrieved documents and explicitly refuse to respond when the retrieved context does not contain relevant information, directly preventing the model from using its parametric knowledge. Groundedness detection validates that the generated completion is actually supported by the provided source material, flagging or blocking responses that contain ungrounded claims — this adds a verification layer even when retrieval succeeds. Setting strictness to 5 controls the relevance threshold for which retrieved documents are passed to the model but does not prevent the model from using its own knowledge when responding. Reducing retrieved documents from 10 to 3 reduces context volume but does not restrict the model from applying general knowledge. Switching to Provisioned throughput affects latency and cost but has no impact on whether the model uses grounded or parametric knowledge.
2. Adatum needs to extract structured data from thousands of invoices monthly. The invoices come from multiple vendors with varying layouts and formats. The solution must identify standard invoice fields like invoice number, date, vendor details, line items, and totals without training custom models. Which Azure service and feature combination meets these requirements? (Select one!)
AAzure AI Vision OCR with custom pattern matching to extract invoice fields
BAzure Document Intelligence prebuilt Invoice model which recognizes standard invoice fields across different formats
CCustom Vision classification model trained on sample invoices from each vendor
DGPT-4 Vision with prompts describing the invoice fields to extract
Explanation
Azure Document Intelligence prebuilt Invoice model is specifically designed to extract standard invoice fields including invoice ID, date, vendor information, customer details, line items, amounts, and taxes from invoices regardless of format variations. It requires no training and handles multiple layouts. Vision OCR provides text extraction but not structured field recognition. Custom Vision is for image classification and object detection, not document understanding. GPT-4 Vision could work but would require careful prompting, cost more per document, and is less optimized for structured extraction than the specialized Invoice model.
3. Contoso is deploying Azure AI Content Safety to analyze user-uploaded images for inappropriate content. The application receives images from mobile apps in various formats. The solution must support all formats natively supported by the service while ensuring images meet size requirements. Which image specifications should you enforce in your application validation logic? (Select two!)
Multiple correct answersAAccept images in JPEG, PNG, GIF, BMP, TIFF, and WEBP formats only
BEnforce a maximum file size of 4 MB per image
CEnforce a maximum file size of 10 MB per image
DValidate that image dimensions are between 50x50 and 7200x7200 pixels
EValidate that image dimensions are between 100x100 and 4096x4096 pixels
Explanation
Azure AI Content Safety supports exactly 6 image formats: JPEG, PNG, GIF, BMP, TIFF, and WEBP. The maximum file size is 4 MB per image. Images must also be at least 50x50 pixels and no more than 7200x7200 pixels. Enforcing the correct format list and the 4 MB limit ensures requests are not rejected by the service. The 10 MB limit is incorrect - Azure AI Content Safety has a strict 4 MB maximum for images. The minimum dimension is 50x50 pixels (not 100x100), and the maximum is 7200x7200 pixels (not 4096x4096). The 100x100 minimum and 4096x4096 maximum apply to different Azure services.
4. Fabrikam is training a CLU model to support multiple languages including Japanese, German, and Spanish. Which training mode must they use? (Select one!)
AStandard training mode supports all languages
BAdvanced training mode is required for non-English languages
CPremium training mode is required for Asian languages
DEither Standard or Advanced mode works for all languages
Explanation
Standard training mode in CLU supports English only. To train models for other languages including Japanese, German, Spanish, or any non-English language, you must use Advanced training mode. This is a key limitation to be aware of when planning multi-lingual applications.
5. An accounting app, 'PayRight', uses an S0 instance of Azure AI Document Intelligence to extract data from client invoices stored as PDF files. The files are all under 10 pages and 2 MB. Users report that some specific invoices consistently fail to process. What is a likely reason for these failures, given that the file size and page count are within limits?
AThe files are too large in size.
BThe files have too many pages.
CThe S0 tier does not support PDF files.
DThe files are password protected.
Explanation
Azure AI Document Intelligence, like many automated processing services, cannot process files that are encrypted or password protected. This is a common security feature that prevents the service from opening the file to read its contents. Since the other limits for the S0 tier (up to 2,000 pages and 500 MB) are well above the file characteristics described, a password-protected PDF is the most probable cause of the issue. The S0 tier absolutely supports PDF files, and the stated page and size limits are not being exceeded.
More Microsoft Practice Exams
Administering Windows Server Hybrid Core Infrastructure (AZ-800)
AZ-800 · 898 questions
Microsoft Certified: Agentic AI Business Solutions Architect (AB-100)
AB-100 · 700 questions
Microsoft Certified: Azure Administrator Associate (AZ-104)
AZ-104 · 757 questions
Microsoft Certified: Azure AI Fundamentals (AI-900)
AI-900 · 464 questions
Microsoft Azure AI Fundamentals (AI-901)
AI-901 · 600 questions
Microsoft Certified: Azure Cosmos DB Developer Specialty (DP-420)
DP-420 · 677 questions
$17.99
One-time access to this exam
1,392 questions (27 practice exams' worth)
Unlimited timed exam simulations
Yours forever — no subscription
Detailed explanations
Free preview stays available