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

Exam Overview

• Certification: Microsoft Azure AI Engineer Associate
• Exam Code: AI-102
• Target Audience: AI engineers building, managing, and deploying AI solutions on Azure
• Experience Required: Python or C# development, REST APIs/SDKs, responsible AI principles

Skills Measured (as of April 30, 2025)

1. Plan and Manage an Azure AI Solution (20-25%)

Select the Appropriate Azure AI Foundry Services

• Select service for generative AI solution
• Select service for computer vision solution
• Select service for natural language processing solution
• Select service for speech solution
• Select service for information extraction solution
• Select service for knowledge mining solution

Plan, Create and Deploy an Azure AI Foundry Service

• Plan for solution meeting Responsible AI principles
• Create Azure AI resource
• Choose appropriate AI models
• Deploy AI models using appropriate deployment options
• Install and utilize SDKs and APIs
• Determine default endpoint for service
• Integrate Azure AI Foundry Services into CI/CD pipeline
• Plan and implement container deployment

Manage, Monitor, and Secure an Azure AI Foundry Service

• Monitor Azure AI resource
• Manage costs for Azure AI Foundry Services
• Manage and protect account keys
• Manage authentication for Azure AI Foundry Service resource

Implement AI Solutions Responsibly

• Implement content moderation solutions
• Configure responsible AI insights (content safety)
• Implement responsible AI (content filters, blocklists)
• Prevent harmful behavior (prompt shields, harm detection)
• Design responsible AI governance framework

2. Implement Generative AI Solutions (15-20%)

Build Generative AI Solutions with Azure AI Foundry

• Plan and prepare for generative AI solution
• Deploy hub, project, and resources with Azure AI Foundry
• Deploy appropriate generative AI model
• Implement prompt flow solution
• Implement RAG pattern by grounding model in data
• Evaluate models and flows
• Integrate project into application with Azure AI Foundry SDK
• Utilize prompt templates

Use Azure OpenAI in Foundry Models

• Provision Azure OpenAI in Foundry Models resource
• Select and deploy Azure OpenAI model
• Submit prompts to generate code and natural language
• Use DALL-E model to generate images
• Integrate Azure OpenAI into applications
• Use large multimodal models
• Implement Azure OpenAI Assistant

Optimize and Operationalize Generative AI Solution

• Configure parameters to control generative behavior
• Configure model monitoring and diagnostics
• Optimize and manage deployment resources
• Enable tracing and collect feedback
• Implement model reflection
• Deploy containers for local and edge devices
• Implement orchestration of multiple generative AI models
• Apply prompt engineering techniques
• Fine-tune generative models

3. Implement an Agentic Solution (5-10%)

Create Custom Agents

• Understand role and use cases of agents
• Configure resources to build agents
• Create agent with Azure AI Foundry Agent Service
• Implement complex agents with Semantic Kernel and Autogen
• Implement complex workflows (orchestration, multi-agent, autonomous)
• Test, optimize and deploy agents

4. Implement Computer Vision Solutions (10-15%)

Analyze Images

• Select visual features for image processing
• Detect objects and generate image tags
• Include image analysis features in processing requests
• Interpret image processing responses
• Extract text from images using Azure AI Vision
• Convert handwritten text

Implement Custom Vision Models

• Choose between image classification and object detection
• Label images
• Train custom image models
• Evaluate custom vision model metrics
• Publish custom vision models
• Consume custom vision models
• Build custom vision models code-first

Analyze Videos

• Use Azure AI Video Indexer for video/live stream insights
• Use Azure AI Vision Spatial Analysis for people detection/movement

5. Implement Natural Language Processing Solutions (15-20%)

Analyze and Translate Text

• Extract key phrases and entities
• Determine text sentiment
• Detect language
• Detect personally identifiable information (PII)
• Translate text and documents using Azure AI Translator

Process and Translate Speech

• Integrate generative AI speaking capabilities
• Implement text-to-speech and speech-to-text
• Improve text-to-speech using SSML
• Implement custom speech solutions
• Implement intent and keyword recognition
• Translate speech-to-speech and speech-to-text

Implement Custom Language Models

• Create intents, entities, and utterances
• Train, evaluate, deploy, and test language understanding models
• Optimize, backup, and recover models
• Consume language models from client applications
• Create custom question answering projects
• Add question-answer pairs and import sources
• Train, test, and publish knowledge bases
• Create multi-turn conversations
• Add alternate phrasing and chit-chat
• Export knowledge bases
• Create multi-language question answering solutions
• Implement custom translation

6. Implement Knowledge Mining and Information Extraction Solutions (15-20%)

Implement Azure AI Search Solution

• Provision Azure AI Search resource
• Create index and define skillset
• Create data sources and indexers
• Implement custom skills in skillsets
• Create and run indexers
• Query index (syntax, sorting, filtering, wildcards)
• Manage Knowledge Store projections

Implement Semantic and Vector Store Solutions

• Configure semantic search
• Implement vector search
• Hybrid search approaches

Implement Azure AI Document Intelligence Solution

• Provision Document Intelligence resource
• Use prebuilt models for data extraction
• Implement custom document intelligence models
• Train, test, and publish custom models
• Create composed document intelligence models

Extract Information with Azure AI Content Understanding

• Create OCR pipeline for text extraction
• Summarize, classify, and detect document attributes
• Extract entities, tables, and images
• Process and ingest various content types

Key Azure AI Services

Azure OpenAI Service

• GPT models (GPT-4, GPT-3.5)
• DALL-E for image generation
• Embeddings models
• Chat completions API
• Function calling

Azure AI Vision

• Image analysis
• OCR (Read API)
• Custom Vision
• Face API
• Video Indexer
• Spatial Analysis

Azure AI Language

• Text Analytics
• Language Understanding (LUIS)
• Question Answering
• Translator
• Custom Named Entity Recognition

Azure AI Speech

• Speech-to-Text
• Text-to-Speech
• Speech Translation
• Speaker Recognition
• Custom Speech

Azure AI Document Intelligence

• Prebuilt models (invoices, receipts, ID)
• Custom extraction models
• Layout API
• General document model

Azure AI Search

• Full-text search
• Semantic search
• Vector search
• AI enrichment
• Knowledge mining

Important Concepts

Responsible AI

• Fairness: Avoid bias
• Reliability & Safety: Consistent performance
• Privacy & Security: Data protection
• Inclusiveness: Accessible to all
• Transparency: Explainable AI
• Accountability: Human oversight

Generative AI Patterns

• RAG (Retrieval Augmented Generation)

  • Ground models in your data
  • Reduce hallucinations
  • Provide context

• Prompt Engineering

  • System messages
  • Few-shot learning
  • Chain of thought
  • Temperature and top-p

Agent Architecture

• Single Agent: One AI agent handling tasks
• Multi-Agent: Multiple specialized agents
• Orchestration: Coordinating agent actions
• Autonomous: Self-directed agents

Development Tools

SDKs

• Azure SDK for Python
• Azure SDK for .NET
• Azure SDK for JavaScript
• Azure SDK for Java

REST APIs

• Authentication (API keys, Azure AD)
• Request/response formats
• Rate limiting
• Error handling

Development Environments

• Azure AI Studio
• Visual Studio Code
• Jupyter Notebooks
• Azure Machine Learning

Best Practices

Security

• Use managed identities
• Rotate API keys regularly
• Implement network isolation
• Enable diagnostic logging
• Use Azure Key Vault

Performance

• Implement caching
• Use batching for bulk operations
• Configure appropriate tier/SKU
• Monitor quotas and limits
• Optimize prompt length

Cost Management

• Choose appropriate pricing tier
• Monitor usage and costs
• Implement caching strategies
• Use commitment tiers for predictable workloads
• Clean up unused resources

Study Resources

Official Microsoft Learn

• AI-102 learning paths
• Azure AI services documentation
• Hands-on labs
• Practice assessments

Hands-On Practice

• Azure free account
• Azure AI Studio
• Sample applications
• GitHub repositories

Community Resources

• Microsoft Q&A forums
• AI/ML Tech Community
• AI Show videos
• Documentation and tutorials

Exam Details

• Passing Score: 700
• Question Format: Multiple choice, case studies, drag-and-drop
• Exam Duration: 120 minutes (150 minutes for non-native English speakers)
• Languages Available: Multiple languages
• Exam Cost: $165 USD (varies by region)

Certification Path

• Prerequisites: Development experience, REST APIs knowledge
• Renewal: Required every 12 months through Microsoft Learn
• Related Certifications:
  • Azure Data Scientist Associate (DP-100)
  • Azure Solutions Architect Expert (AZ-305)
  • Azure Developer Associate (AZ-204)