Azure RAG AI Template

This template helps you build Retrieval-Augmented Generation (RAG) AI applications using Azure OpenAI and Azure AI Search. Follow the steps below to understand RAG, set up Azure, and run the code or API.

---

1. What is Retrieval-Augmented Generation (RAG)?

RAG combines information retrieval and AI generation. When a user asks a question, the system:

1. Retrieves relevant documents from a large data source (like PDFs or images in Azure Blob Storage) using Azure Cognitive Search and vector embeddings.
2. Uses Azure OpenAI (e.g., GPT-4) to generate a response based on both the user’s question and the retrieved documents.

This approach gives more accurate, context-aware answers by grounding AI responses in your own data.

---

2. Azure RAG AI Architecture Diagram

How it works:

1. User / Application: The user sends a message (text or image) to the application.
2. OpenAI gpt-4o-mini: The message is processed by the GPT-4o-mini model, which can use context from the vector data (retrieved documents) via AI Search.
3. AI Search with Vector Index: The system searches for relevant content using a vector index built from your documents. This index is created using embeddings from the OpenAI text-embedding-ada-002 model.
4. OpenAI text-embedding-ada-002: This model generates vector embeddings for your documents, making them searchable by semantic meaning.
5. Blob Storage: All your source documents (PDFs, images, Excel files, etc.) are stored here. The embedding model processes these files to create the vector index.

The flow: User → GPT-4o-mini → (AI Search) → Vector Index (built from Blob Storage via embeddings) → Response to User

---

3. How to Configure Azure

You’ll need:

• An Azure subscription
• Azure OpenAI resource (with GPT-4o-mini and text-embedding-ada-002 models)
• Azure Blob Storage (for your documents)
• Azure Cognitive Search (for vector search)

Step-by-step setup:

1. Create Azure OpenAI Resource

   • In the Azure portal, create an OpenAI resource in the East US region.
   • Deploy the gpt-4o-mini model (supports text and image input).
   • Deploy the text-embedding-ada-002 model for vector embeddings.

2. Create Storage Account

   • Create a Storage Account (Standard, LRS) in the same region.
   • In the storage account, create a container and upload your data (e.g., brochures.zip from this link).

3. Create Azure Cognitive Search Resource

   • Create a Cognitive Search resource (Free tier is fine) in the same region.
   • In the Cognitive Search portal, use “Import and vectorize data” to connect to your Blob Storage and set up a vector index using the text-embedding-ada-002 model.

4. Configure Keys and Endpoints

   • In the Azure portal, copy the following values:
     • OpenAI: KEY1, Endpoint, and deployment name
     • Cognitive Search: Url, Primary admin key, and index name
   • Paste these into the appsettings.json file in both api/ and simple-code/ folders:
     • AzureOAIKey, AzureOAIEndpoint, AzureOAIDeploymentName
     • AzureSearchEndpoint, AzureSearchKey, AzureSearchIndex

---

4. How to Run the Simple Code Example

The simple-code/ folder contains a minimal example (RAGAI.cs) showing how to use Azure OpenAI and Cognitive Search together.

Steps:

1. Open a terminal and navigate to the folder:

   cd simple-code

2. Restore dependencies:

   dotnet restore

3. Build the project:

   dotnet build

4. Run the example:

   dotnet run

---

5. How to Run the REST API

The api/ folder contains a REST API with endpoints for chat and RAG operations.

Steps:

1. Navigate to the API folder:

   cd api

2. Restore dependencies:

   dotnet restore

3. Build the API:

   dotnet build

4. Run the API:

   dotnet run

5. Access the API documentation at:
   • https://localhost:7222/swagger/index.html

Main Endpoints:

• POST /api/v1/AzureOpenAI/chat — Text-based RAG chat
• POST /api/v1/AzureOpenAI/chat-with-image — Image-based RAG chat

---

6. How to Run the Unit Tests

Unit and integration tests are in the api.Tests/ folder.

Steps:

1. Navigate to the test folder:

   cd api.Tests

2. Run the tests:

   dotnet test

---

Additional Features

• SOLID architecture for maintainability
• Rate limiting and validation
• In-memory mode for local development
• Swagger UI for API exploration

For more details, see the code comments and the architecture diagram in image/azure-rag-ai-diagram.png.