Microsoft, certification course, intermediate level
AI-103: Build and Ship AI Agents on Microsoft Azure
Build AI apps and agents in Python on Microsoft Foundry, ship one to production, and prepare for exam AI-103
Free. You sign in once to start.
- 6
- modules
- 164
- video lessons
- 10
- quizzes
- 8.9 h
- of video
- 142 lectures
- 22 demos
- 10 quizzes
Each bar is one lesson. Gaps separate the modules.
What you will be able to do
- Build a Python chat app on Microsoft Foundry with the Responses API, tools, file search grounding and responsible AI guardrails
- Create AI agents that take action with custom function tools, MCP servers and Foundry IQ knowledge bases
- Orchestrate workflows and multi-agent teams in Python with the Microsoft Agent Framework, and trace them in Application Insights
- Add text analysis, speech, real-time voice with Voice Live and translation with Azure Language, Azure Speech and Azure Translator
- Turn images, documents, audio and video into structured data with vision models, Content Understanding and Azure AI Search
- Ship an agent to production as a Foundry hosted agent with its own identity, a toolbox, a web front end and GitHub Actions CI/CD
- Secure AI apps with keyless Microsoft Entra ID sign-in, and monitor, trace and roll back agents in production
- Prepare for Microsoft exam AI-103, Developing AI Apps and Agents on Azure, across all five skill domains
Curriculum
174 items (lessons and quizzes) across 6 modules. Open a module to see its lessons.
Course Introduction: The Meridian Story and the AI-103 Exam1 lesson
- Welcome: Course Roadmap, the Meridian Story and the AI-103 Exam
Module 1: Build Generative AI Apps on Microsoft Foundry35 lessons
- Module 1 Introduction: Meridian's First Chat Assistant on Foundry
- Demo: Get Your Azure Subscription
- Demo: AI Services on Azure
- Demo: Microsoft Foundry on Azure
- Demo: A Model, End to End
- Setup: Install Python 3.13
- Setup: Install VS Code and Run Python
- Lab Setup: Overview and Setup
- Meet Microsoft Foundry: Resources, Projects, Foundry Tools and SDKs
- Endpoints, Keyless Identity, Foundry Roles and Responsible AI Principles
- Choosing a Model: The Foundry Model Catalog and Benchmarks
- Deploying and Evaluating Models, Plus Lab Part 1: Project and First Call
- Lab Part 1.1: Deploy the Model
- Lab Part 1.2: The First Call
- From Playground to Code: View Code, Endpoints and SDK Choice, Lab Part 2
- Lab Part 2.1: The Persona in the Playground
- Lab Part 2.2: The Same Rules From Code
- Two Chat APIs: Chat Completions vs Responses API in Python, Lab Part 3
- Lab Part 3.1: Chat Loop with the Responses API
- Lab Part 3.2: The Chat Completions Way
- Module 1 Quiz 1: Sections 1.01 to 1.06
- Giving the Model Tools: Code Interpreter and Web Search, Lab Part 4
- Lab Part 4.1: code_interpreter
- Lab Part 4.2: web_search
- Grounding and Functions: File Search and Function Calling, Lab Part 5
- Lab Part 5.1: Policy PDFs in a Vector Store
- Lab Part 5.2: The Booking Lookup
- Optimizing Behavior: Prompt Engineering, RAG or Fine-Tuning, Lab Part 6
- Lab Part 6.1: A Better System Message
- Lab Part 6.2: Test v1 Against v2
- Guardrails: Responsible Generative AI and Prompt Shields, Lab Part 7
- Lab Part 7.1: A Stricter Guardrail
- Lab Part 7.2: The Attack Test
- Module 1 Quiz 2: Sections 1.07 to 1.10
- Module 1 Wrap-up: Lab Complete, Keep Current and What's Next
Module 2: Build AI Agents That Act37 lessons
- Module 2 Introduction: From Assistant to Agent - Road Map and Lab
- Lab Setup: Overview and Setup
- Agents in Foundry: What an Agent Is and Your First Agent in Python
- Lab Part 1.1: The Agent in the Portal
- Lab Part 1.2: The Agent in Code
- Custom Tools: Function Calling, Azure Functions, OpenAPI and Logic Apps
- Lab Part 2.1: Describe the Tools
- Lab Part 2.2: Run the Tools
- The Model Context Protocol: Connect MCP Servers to Your Foundry Agent
- Lab Part 3.1: The Partner Hotels MCP Server
- Lab Part 3.2: A Tunnel to the Server
- Lab Part 3.3: The Agent Uses MCP
- Foundry IQ: Ground Your Agent with RAG, Knowledge Bases and Citations
- Demo: Agent Tools in the Portal
- Demo: An MCP Tool From the Catalog
- Lab Part 4.1: Connect Foundry IQ
- Lab Part 4.2: The Policy Knowledge Base
- Demo: Foundry IQ Knowledge
- Module 2 Quiz 1: Sections 2.01 to 2.04
- Microsoft 365: Publish Your Agent to Teams and Copilot, and Use Work IQ
- Agent Workflows: Executors, Edges and Events with the Agent Framework
- Demo: Publish an Agent
- Lab Part 5.1: Three Agents for Refunds
- Lab Part 5.2: Wire, Run and Approve
- The Microsoft Agent Framework: Code-First Agents with Tools and Sessions
- Lab Part 6.1: Tools as Python Functions
- Lab Part 6.2: The Agent and Its Session
- Multi-Agent Orchestration: Five Patterns and a Sequential Trip Planner
- Lab Part 7.1: Specialist Agents
- Lab Part 7.2: Run the Team
- Agents in Production: Lifecycle, Hosted Agents, Memory and Observability
- Lab Part 8.1: Connect Application Insights
- Lab Part 8.2: Traces From Code
- Lab Part 8.3: Read the Trace
- Demo: Evaluate an Agent
- Module 2 Quiz 2: Sections 2.05 to 2.09
- Module 2 Wrap-up: Lab Checklist and Platform Changes to Keep Current
Module 3: Natural Language: Text, Speech, Voice and Translation38 lessons
- Module 3 Introduction: Give the Agent Ears and a Voice
- Lab Setup: Overview and Setup
- Analyze Text with Azure Language: Language Detection, NER and PII Redaction
- Lab Part 1.1: The Language Client
- Lab Part 1.2: Run the Analysis
- Demo: The Language Playground
- The Language MCP Server: Text Analysis Tools for Your Foundry Agent
- Lab Part 2.1: Connect the Foundry Resource
- Lab Part 2.2: The Language MCP Tool
- Lab Part 2.3: A Second Message
- Demo: The Language MCP Tool in the Portal
- Speech-Capable Models: gpt-4o-transcribe and gpt-4o-mini-tts in Foundry
- Lab Part 3.1: The Audio Models
- Lab Part 3.2: Transcribe and Reply
- Lab Part 3.3: Listen and Tune the Voice
- Build Speech Apps with Azure Speech: Speech to Text, Text to Speech and SSML
- Lab Part 4.1: Ears and Voice for the Kiosk
- Lab Part 4.2: Talk to the Kiosk
- Lab Part 4.3: SSML and Portuguese
- Demo: The Speech Playground
- Module 3 Quiz 1: Sections 3.01 to 3.04
- The Speech MCP Server: Audio Tools for Your Agent via Blob Storage
- Lab Part 5.1: Storage for the Audio
- Lab Part 5.2: Add the Speech MCP Server
- Lab Part 5.3: The Voice Agent
- Voice Live: Build Real-Time Speech-to-Speech Voice Agents
- Lab Part 6.1: The Voice Live Playground
- Lab Part 6.2: The Phone Agent in Code
- Lab Part 6.3: Make the Call
- Demo: Voice Mode for an Agent
- Translate Text and Speech: Azure Translator, Speech SDK and Generative Models
- Lab Part 7.1: Translate the Messages
- Lab Part 7.2: Translate Speech
- Lab Part 7.3: Wire It Together
- Demo: Translator in the Portal
- Demo: Customize with Your Own Data
- Module 3 Quiz 2: Sections 3.04 to 3.07
- Module 3 Wrap-up: Lab Recap and Platform Changes to Keep Current
Module 4: Visual Data: Vision, Media, Content Understanding, Search36 lessons
- Module 4 Introduction: Extract Insights from Visual Data in Azure
- Lab Setup: Overview and Setup
- Build a Vision-Enabled App: Send Images to Multimodal Models in Python
- Lab Part 1.1: Generate the Sample Photos
- Lab Part 1.2: Describe the Damage
- Lab Part 1.3: Compare with Chat Completions
- Demo: Vision in the Playground
- Generate Images and Video: Images API, Video Prompts and the Video Job
- Lab Part 2.1: Deploy the Image and Video Models
- Lab Part 2.2: The Hero Image
- Lab Part 2.3: The Teaser Video
- Demo: The Image Playground
- Demo: The Video Playground
- Analyze Images with Content Understanding: Analyzers, Schemas and the SDK
- Lab Part 3.1: Default Models for Content Understanding
- Lab Part 3.2: Build the Photo Analyzer
- Lab Part 3.3: Tag the Hotel Photos
- Lab Part 3.4: Add a Field
- Module 4 Quiz 1: Sections 4.01 to 4.03
- Documents, Audio and Video with Content Understanding: Custom Analyzers
- Lab Part 4.1: The Invoice Analyzer
- Lab Part 4.2: Pay or Review
- Lab Part 4.3: The Support Call
- Lab Part 4.4: Analyze the Call
- Demo: Content Understanding in the Portal
- Demo: Audio and Video in the Portal
- Knowledge Mining with Azure AI Search: Indexers, Skillsets, Knowledge Store
- Lab Part 5.1: Upload the Archive
- Lab Part 5.2: The Search Service Identity
- Lab Part 5.3: Build the Index
- Lab Part 5.4: Query the Archive
- Lab Part 5.5: The Archive Agent
- Demo: The Search Pipeline in the Portal
- Demo: Query the Archive in the Portal
- Module 4 Quiz 2: Sections 4.04 to 4.05
- Module 4 Wrap-up: Lab Recap, Platform Changes and What Comes Next
Module 5: Capstone: Ship Meridian's Agent to Production27 lessons
- Capstone Introduction: Ship Meridian's Agent to Production
- Lab Setup: Overview and Setup
- Lab Setup 0.1: Provision the Resources
- Lab Setup 0.2: The Archive Index
- From Laptop to Hosted Agent: Prompt Agents vs Hosted Agents
- Lab Part 1.1: The Agent on the Laptop
- Lab Part 1.2: Talk to It
- Package and Deploy: Container Image, ACR and the First Agent Version
- Lab Part 2.1: Build the Image
- Lab Part 2.2: Create the Hosted Agent
- Lab Part 2.3: The Agent in the Portal
- Protocols and Sessions: Responses, Invocations and WebSocket
- Lab Part 3.1: Talk to the Endpoint
- Lab Part 3.2: Traces and Sessions
- Module 5 Quiz 1: Sections 5.01 to 5.03
- Identity, Secrets and Tools: Agent Identity, Key Vault and Toolbox
- Lab Part 4.1: The Agent's Own Identity
- Lab Part 4.2: One Toolbox
- Lab Part 4.3: Version Two
- Front End and Pipeline: App Service, GitHub Actions and Rollback
- Lab Part 5.1: A Page for Travelers
- Lab Part 5.2: Sign in From GitHub Without Secrets
- Lab Part 5.3: Ship a Change with a Push
- Lab Part 5.4: Traces and Rollback
- Module 5 Quiz 2: Sections 5.04 to 5.05
- Capstone Wrap-up: Platform Changes to Watch and Course Conclusion
- Bonus Lecture: Congratulations and Your Next Course
About this course
Most AI courses teach Azure services one at a time. This one tells one story: you build a single AI agent, module by module, and ship it to production.
You join Meridian Travel Group, a travel company whose contact center answers the same questions all day, whose policies and invoices sit in an archive, and whose customers write and call in many languages. Every module adds new skills to the same agent, so by the end you have built something real, not a set of disconnected demos.
What you build
Module 1, Generative AI apps
a Python chat app on Microsoft Foundry. You choose, deploy and evaluate models, call the Responses API, add code interpreter, web search, file search and function calling, tune the system message, and protect the app with guardrails.
Module 2, AI agents
the app becomes an agent that acts, with custom tools, an MCP server, Foundry IQ knowledge, workflows, the Microsoft Agent Framework, a multi-agent team and tracing in Application Insights.
Module 3, Natural language
text analysis with Azure Language, speech apps with Azure Speech, the Language and Speech MCP servers, a real-time phone agent with Voice Live, and translation with Azure Translator.
Module 4, Visual data
vision models, image and video generation, Content Understanding for photos, invoices and calls, and knowledge mining over a document archive with Azure AI Search.
Module 5, Capstone
the agent goes to production as a Foundry hosted agent with its own identity, one toolbox for every tool, a web page for travelers, and a GitHub Actions pipeline that builds and deploys every change, with traces and a one-step rollback.
How every module works
Story
A short chapter sets up the problem
Concept
Built up on screen as a diagram
Code
Real code with the current SDK names
Lab
Every step run on camera
Demo
The same idea in the portal
A short story chapter sets up Meridian's problem. The concept builds up step by step on screen. You see real Python code with the current SDK names. Then come the hands-on lab videos, where every line is typed and run on camera, and short portal demos that show the same features with no code.
What you get
8.9
hours of video
142
lectures
22
portal demos
10
quizzes
- 8.9 hours of video: 48 lectures, 92 lab videos, 22 portal demos and 2 setup videos
- A complete lab in every module, built in parts that join into one working application
- the full lab code and sample data, lab guides, demo guides with screenshots, and the slides of every section (PDF)
- 10 quizzes with 100 exam-style questions on concepts, code and the portal, plus a knowledge check for Modules 1 to 4
- English captions on every video
Exam AI-103 and beyond
The course covers all five skill domains of Microsoft exam AI-103, Developing AI Apps and Agents on Azure, which leads to the Microsoft Certified: Azure AI App and Agent Developer Associate certification. Wherever Microsoft is retiring a feature, you learn its successor, and every module closes with a Keep current slide. The capstone then goes further than the exam and takes the agent all the way to production, the way you would do it at work.
By the end, you can design, build, secure and ship AI apps and agents on Azure, and you are ready to prepare for exam day with confidence.