🧠 Agentic RAG Service using .NET 8, Azure OpenAI & Semantic Kernel

This repository demonstrates how to build a Retrieval-Augmented Generation (RAG) service using Microsoft Semantic Kernel SDK, .NET 8, and Azure AI Search. It includes both Traditional RAG and Agentic RAG implementations with structured prompt execution and dynamic agent reasoning.

---

📌 Features

• 🔎 Traditional RAG with Azure Cognitive Search and contextual prompt injection
• 🤖 Agentic RAG using Semantic Kernel's ChatCompletionAgent and AgentThread
• 🌦 Weather plugin integration (example)
• 💡 Modular architecture with plugin support and step-by-step reasoning

---

---

🚀 Getting Started

🔧 Configuration

Before running the project, you need to replace placeholder values with your actual Azure credential in launchSettings.json

      "environmentVariables": {
        "ASPNETCORE_ENVIRONMENT": "Development",
        "AZURE_OPENAI_URL": "https://<RESOURCENAME>.openai.azure.com",
        "AZURE_OPENAI_KEY": "<ADD_KEY>",
        "AZURE_AI_SEARCH_URL": "https://<RESOURCENAME>.search.windows.net",
        "AZURE_AI_SEARCH_KEY": "<ADD_KEY>",
        "EMBEDDING_DEPLOYMENT_NAME": "<REPLACE_WITH_DEPLOYMENT_NAME>",
        "GPT_DEPLOYMENT_NAME": "<LLM_MODEL_DEPLOYMENT_NAME>",
        "INDEX_NAME": "<AI_SEARCH_INDEX_NAME>"
      },

Also, ensure your Azure AI Search setup and embedding model configurations are correctly referenced in your AzureSearchService.cs.

---

▶️ How to Run

📌 Prerequisites

• .NET 8 SDK
• Valid Azure OpenAI and Azure AI Search access
• Optional: Visual Studio or VS Code for debugging and running

🛠 Build and Run the Project

You can run the service locally using the .NET CLI:

git clone https://github.com/your-username/AgenticRAG.git
cd AgenticRAG
dotnet restore
dotnet build
dotnet run

This will start the web API and allow you to access endpoints that trigger traditional and agentic RAG behavior.

---

📬 API Endpoints (Example)

Note: Update based on actual controller methods in your project

POST /rag/traditional
{
  "query": "Find best hotels in Sedona"
}

POST /rag/agentic
{
  "prompt": "Plan a weekend trip for family to Acadia National Park"
}

🚀 Usage

🔹 Traditional RAG

var result = await ragService.GenerateResponseUsingTraditionalRag("Best hotels in Bar Harbor?");
Console.WriteLine(result);

This version:

• Embeds the user query
• Retrieves relevant documents using Azure AI Search
• Constructs a prompt using retrieved context
• Sends it to the LLM for response generation

---

🔹 Agentic RAG

var result = await ragService.GenerateResponseUsingAgenticRag("What's the weather like for a trip to Acadia National Park?");
Console.WriteLine(result);

This version:

• Breaks down the prompt into sub-tasks
• Optimizes and refines search queries
• Invokes tools (e.g., weather)
• Reflects and reasons on the context before finalizing the answer

The agent returns:

• Final answer
• Reasoning trace
• Tools and context used

---

🧩 Plugins

Example Weather Plugin added as:

kernel.ImportPluginFromType<WeatherPlugin>("searchWeather");

You can build and register your own plugins to expand capabilities.

---

📚 Technologies Used

• .NET 8
• Microsoft Semantic Kernel
• Azure OpenAI
• Azure Cognitive Search

---