Building .NET AI Applications with Semantic Kernel and Amazon Bedrock

Introduction to Agentic AI

What is an Agent?

• An agent is an LLM (or other model type) that exists in a loop, able to call on tools to do additional work
• This allows the LLM to extend its functionality and capabilities beyond its initial training
• Agents are a philosophical concept more than a specific technology, focusing on the loop and tool-calling abilities

Real-World Applications of Agentic AI

• Summarizing meeting notes and extracting action items
• Personalization in applications like e-commerce
• Building support chatbots that can interact with API documentation

Microsoft Frameworks for Building .NET AI Applications

Microsoft.Extensions.AI

• Provides a simple abstraction layer for interacting with LLMs, hiding the complexities of different providers
• Allows building a generic "chat client" that can be backed by different LLM providers

Semantic Kernel

• Framework built on top of Microsoft.Extensions.AI
• Provides plugins (functions/tools), AI models, hooks (middleware), and filters (permission-based hooks)
• Allows easily building the "agent loop" with tool-calling capabilities

Microsoft Agent Framework

• Newer framework from Microsoft, simpler than Semantic Kernel
• Provides the agent loop, tool-calling, chat history, and other core agent capabilities
• Does not yet have the same level of features and extensibility as Semantic Kernel

AWS Tools for Running .NET AI Applications

Amazon Bedrock

• AWS cloud service providing pre-trained LLMs and optimization for application developers
• Offers a more application-focused alternative to training models in SageMaker

Amazon Agent Core

• Serverless infrastructure for running LLM applications, including:
  • Runtime: Containerized execution of the LLM application
  • Gateway: Provides access to other AWS services like Lambda
  • Code Interpreter: Isolated sandbox for running code snippets
  • Memory: Provides short-term and long-term memory for the agent
  • Observability: Integrated logging and monitoring

Integrating AWS and Microsoft Frameworks

• Semantic Kernel has native integration with Amazon Bedrock
• Microsoft Agent Framework currently lacks built-in support for AWS services, requiring custom integration
• Example project demonstrates using both Semantic Kernel and Microsoft Agent Framework with Amazon Agent Core

Demonstration and Key Takeaways

Semantic Kernel Example

• Implements a "horoscope agent" using Semantic Kernel
• Provides daily horoscope functionality, but lacks the ability to retrieve weekly or monthly horoscopes

Microsoft Agent Framework Example

• Also implements a "horoscope agent" using Microsoft Agent Framework
• Leverages Amazon Agent Core Gateway to expose horoscope API endpoints as MCP tools
• Able to retrieve daily, weekly, and monthly horoscopes

Key Takeaways

• Semantic Kernel provides more features and extensibility, but Microsoft Agent Framework is simpler to use
• Integrating AWS services like Agent Core requires custom work with Microsoft Agent Framework, unlike the native Bedrock support in Semantic Kernel
• The presenters encourage feedback and contributions to improve the .NET experience for these AI frameworks and AWS services

Contact Information

• Nikki: [email]nikki@example.com[/email], @nikkigithub
• AM: [email]am@example.com[/email], @amgithub