A hands-on workshop exploring Microsoft's Semantic Kernel framework for building intelligent AI applications. This workshop provides practical experience with real-world AI application patterns using Python and Azure OpenAI.
This workshop takes you from foundational concepts to advanced implementation patterns through a series of Jupyter notebooks and practical examples. You'll learn how to:
- Build AI applications using Microsoft's Semantic Kernel framework
- Create and orchestrate AI agents with different capabilities and roles
- Construct structured AI workflows using the Process Framework
- Implement enterprise-ready AI features with security and scalability in mind
Experience Semantic Kernel in action through our interactive playground! This visual demonstration allows you to directly engage with the core concepts covered in the workshop.
The playground offers a hands-on environment where you can:
- Test semantic functions in real-time
- Explore agent capabilities and interactions
- Experiment with memory and embeddings
- Try out native plugin integration
- See the Process Framework in action
No need to wait until the end of the workshop - you can start exploring the playground at any time to reinforce concepts as you learn them!
For setup instructions and details on how to run the playground, refer to the Playground README.
- Python 3.10 or higher
- Azure OpenAI API access (API key, endpoint, and deployment name)
- Basic knowledge of Python programming
- Understanding of prompt engineering concepts (helpful but not required)
- UV package manager
The project is managed by pyproject.toml and uv package manager.
For local execution init the .venv environment using uv package manager:
uv sync --prerelease=allow
. ./.venv/bin/activateOBS! At the time of writing the workshop depends on the prerelease libraries.
-
Clone this repository
-
Create a virtual environment:
Linux/macOS:
# Create a virtual environment python -m venv venv # Activate the virtual environment source venv/bin/activate
Windows:
# Create a virtual environment python -m venv venv # Activate the virtual environment venv\Scripts\activate
-
Copy the environment variables template:
cp .env.example .env
-
Add your Azure OpenAI credentials to the
.envfile:AZURE_OPENAI_ENDPOINT=https://xxxxxx.openai.azure.com/ AZURE_OPENAI_CHAT_DEPLOYMENT_NAME=gpt-4o AZURE_OPENAI_EMBEDDING_DEPLOYMENT_NAME=text-embedding-ada-002 AZURE_OPENAI_API_KEY=xxxxxxxxxxx AZURE_OPENAI_API_VERSION=2025-03-01-preview -
Start with the first notebook:
- Begin with
01-intro-to-semantic-kernel/01-intro.ipynb, which includes instructions for installing Semantic Kernel and other required packages.
- Begin with
Learn the fundamentals of Semantic Kernel:
- Core architectural components (Kernel, AI Services, Plugins)
- Building semantic functions with prompts
- Creating native functions with Python code
- Enabling automatic function calling for AI agents
Key Notebooks:
01-intro.ipynb: Core concepts, services, and function creation
Master the creation and orchestration of AI agents:
- Creating specialized agents with different personas
- Implementing multi-agent communication patterns
- Agent selection strategies and orchestration
- Building agent topologies for complex scenarios
- Integrating plugins with agents for enhanced capabilities
Key Notebooks:
02.1-agents.ipynb: Creating and configuring agents02.2-agents-chats.ipynb: Inter-agent communication and complex patterns
Learn to how to connect an SK Agent to MCP:
- Running your MCP server
- Using an Agent in Semantic Kernel to make calls to it
Key Notebooks:
03.1-sk-with-mcp.ipynb: Semantic Kernel with MCP example
Learn to build structured, event-driven AI workflows:
- Understanding the Process Framework architecture
- Defining events, steps, and state management
- Building conversational AI systems with processes
- Implementing complex business logic with AI capabilities
- Creating maintainable and testable AI workflows
Key Notebooks:
04.1-intro-to-processes.ipynb: Building stateful, event-driven AI processes
Master advanced agent communication patterns:
- Implementing Agent-to-Agent (A2A) communication protocols
- Creating specialized agents that can communicate with each other
- Building distributed AI systems with modular, reusable agents
- Understanding A2A server components and agent cards
- Developing production-ready agent architectures
Key Notebooks:
05-semantic-kernel-a2a-tutorial.ipynb: Complete A2A implementation with flight booking and travel planning agents
Standalone Applications:
flight-booking-agent/: Specialized flight booking agent servertravel-booking-agent/: Travel planning agent with web interface
semantic-kernel-workshop/
├── 01-intro-to-semantic-kernel/ # Introduction to core concepts
│ └── 01-intro.ipynb # Basic concepts and functions
├── 02-semantic-kernel-agents/ # Agent creation and orchestration
│ ├── 02.1-single-agents.ipynb # Agent fundamentals
│ ├── 02.2-agents-chats.ipynb # Multi-agent communication
├── 03-semantic-kernel-mcp/ # Using SK with MCP
│ └── 03.1-sk-with-mcp.ipynb # SK + MCP example
├── 04-process-framework/ # Structured AI workflows
│ └── 04.1-intro-to-processes.ipynb # Process fundamentals
├── 05-semantic-kernel-a2a/ # Agent-to-Agent communication
│ ├── 05-semantic-kernel-a2a-tutorial.ipynb # Complete A2A tutorial
│ ├── flight-booking-agent/ # Standalone flight booking agent
│ └── travel-booking-agent/ # Standalone travel planning agent
├── playground/ # Interactive application
│ ├── backend/ # FastAPI server
│ ├── frontend/ # React application
│ ├── start.sh # Launch script
│ └── README.md # Playground documentation
└── .env.example # Environment variables template
For optimal learning, follow the repository's folders in numerical order.
For advanced patterns and enterprise deployment scenarios, explore the Semantic Kernel Advanced Usage repository, which includes:
- Dapr integration for scalable, distributed systems
- Authentication and security patterns
- Natural language to SQL conversion
- Copilot Studio integration
- Microsoft Graph API integration
- Production deployment architecture
This project is licensed under the MIT License - see the LICENSE file for details.