The Recipe Chat solution is an exploration of Semantic Kernel's Agent Group Chat that aims to create and adjust food recipes using multiple Agents in a group chat. There are two user experiences that can be run: Console App and a Blazor Web App.

The solution is divided into three main projects:

1. RecipeChat.Console: A console-based application for interacting with the Recipe Chat functionality.
2. RecipeChat.Web: A web-based application with a Blazor front-end and SignalR used to create a real-time chat experience utilizing the Recipe Chat functionality.
3. RecipeChat.GroupChat: The library that contains all of the SK AgentGroupChat logic including prompt templates for all of the agents as well as selection and termination strategies.

• Purpose: Provides a console-based interface for interacting with the Recipe Chat functionality.
• Key Files:
  • Program.cs: Entry point for the console application.
  • Worker.cs: Handles the main chat loop and integrates with the RecipeGroupChat library.
  • .env: Configuration file for environment variables (e.g., Azure OpenAI endpoint).
• Technologies:
  • .NET Worker Service
  • OpenTelemetry for logging and tracing

• Purpose: A web-based interface for interacting with the Recipe Chat functionality.
• Key Files:
  • Program.cs: Configures the web application and SignalR hub.
  • RecipeChat.Web.Client/Pages/Chat.razor: The main chat page.
  • RecipeChat.Web/Hubs/ChatHub.cs: SignalR hub for real-time communication.
• Technologies:
  • Blazor Web App (utilizing both interactive and server-side components)
  • SignalR for real-time communication

• Purpose: Contains the core logic for utilizing Semantic Kernel to manage and execute the group chat.
• Key Files:
  • RecipeGroupChat.cs: Contains all of the group chat logic.
  • PromptTemplates/: YAML files containing definitions for each Agent as well as the selection and termination strategies that are used to drive the group chat.
• Technologies:
  • Semantic Kernel Agent Framework

• .NET 9.0 SDK
• Azure CLI
• Azure OpenAI Service instance
• Both the Console App and Web App utilize DefaultAzureCredential and assume you are logged in via the Azure CLI with az login and have the Cognitive Services OpenAI User RBAC role granted to you on the AOAI instance.

Create a .env file in the RecipeChat.Console project with the following variables:

AZURE_OPENAI_ENDPOINT="https://[YOUR_ENDPOINT_NAME].openai.azure.com"
AZURE_OPENAI_DEPLOYMENT_NAME="[YOUR_MODEL_DEPLOYMENT_NAME]"
OTEL_ENDPOINT="[YOUR_OTEL_ENDPOINT]"
LOG_HTTP_REQUESTS="true"

Populate the two settings in appsettings.json in the RecipeChat.Web project:

  "AZURE_OPENAI_ENDPOINT": "https://[YOUR_ENDPOINT_NAME].openai.azure.com",
  "AZURE_OPENAI_MODEL_NAME": "[YOUR_MODEL_DEPLOYMENT_NAME]"

There are VS Code launch configurations for both the console and web applications. Utilize these to run your desired experience.

Whether you are running the console app or the web app, the group chat experience is meant to be a multi-turn conversation between the user and the group chat agents. Considering that, here is a simple prompt to try:

User: I'm hungry! Can you make me a recipe?

[Agent Response...]

User: I forgot to mention that I am vegan and have a gluten allergy.

[Agent Response that will call upon the VeganReviewerAgent and GlutenFreeReviewerAgent...]