AgenticRAG — Agentic RAG with Semantic Kernel + Azure AI Search

Overview

AgenticRAG demonstrates an agentic Retrieval-Augmented Generation (RAG) pattern using Microsoft Semantic Kernel and Azure AI Search. The service exposes simple HTTP endpoints that accept a user prompt and return answers produced either by a "traditional" RAG prompt or by an agent-driven RAG flow.

Key Components

• API: Minimal ASP.NET Web API in AgenticRAG/AgenticRAG/Program.cs and AgenticRAG/AgenticRAG/Controllers/RagController.cs.
• RAG service: RagService implements both GenerateResponseUsingTraditionalRag and GenerateResponseUsingAgenticRag (AgenticRAG/AgenticRAG/Services/RagService/RagService.cs).
• Vector search: AzureSearchService wraps Azure Cognitive Search + OpenAI embeddings for vector retrieval.
• Plugin: Example WeatherPlugin shows how to import functionality into the kernel.

Features

• Traditional RAG: fetches top context from Azure AI Search and prompts the model.
• Agentic RAG: uses Semantic Kernel agents for multi-step reasoning and tool invocation.
• CORS and static file hosting for quick front-end experiments.

Environment variables

Set these environment variables before running the app (names come from AgenticRAG/AgenticRAG/Constants/EnvVariables.cs):

• AZURE_OPENAI_URL — Azure OpenAI endpoint URL
• AZURE_OPENAI_KEY — Azure OpenAI key
• AZURE_AI_SEARCH_URL — Azure Cognitive Search endpoint
• AZURE_AI_SEARCH_KEY — Azure Cognitive Search key
• EMBEDDING_DEPLOYMENT_NAME — embedding deployment name
• INDEX_NAME — search index name
• GPT_DEPLOYMENT_NAME — chat / completion deployment name

Quickstart (local)

1. Open a terminal and set the environment variables (PowerShell example):

$env:AZURE_OPENAI_URL="https://your-openai-endpoint"
$env:AZURE_OPENAI_KEY="your-openai-key"
$env:AZURE_AI_SEARCH_URL="https://your-search-endpoint"
$env:AZURE_AI_SEARCH_KEY="your-search-key"
$env:EMBEDDING_DEPLOYMENT_NAME="your-embedding-deployment"
$env:INDEX_NAME="your-index-name"
$env:GPT_DEPLOYMENT_NAME="your-gpt-deployment"

2. Restore and run the API from the solution folder:

cd AgenticRAG
dotnet restore
dotnet run --project AgenticRAG/AgenticRAG.csproj

The API runs on the configured Kestrel port (HTTPS). In development the Swagger UI is enabled.

API Endpoints

• POST /traditionalRag — body: { "prompt": "..." } — returns a JSON Response with Content containing the answer.
• POST /agenticRag — body: { "prompt": "..." } — runs the agentic RAG flow and returns Content.

Example curl (replace host and port):

curl -X POST https://localhost:5001/traditionalRag \
	-H "Content-Type: application/json" \
	-d '{"prompt":"Plan a 3-day trip to Paris focused on museums"}'

Where to look in the code

• Startup and DI: AgenticRAG/AgenticRAG/Program.cs
• Controller: AgenticRAG/AgenticRAG/Controllers/RagController.cs
• RAG implementation: AgenticRAG/AgenticRAG/Services/RagService/RagService.cs
• Search integration: AgenticRAG/AgenticRAG/Services/SearchService.cs

Notes and next steps

• Replace the WeatherPlugin stub with real tool integrations as needed.
• Add index population scripts for Azure Cognitive Search if you want reproducible demos.

---