python

Agentic RAG with Azure AI Search - Python Starter

This is the starter project for Challenge 11 - Build Agentic RAG with Azure AI Search (Python version).

Prerequisites

• Python 3.8 or higher
• Azure AI Search service (Standard tier or higher)
• Azure OpenAI service with deployed models:
  • Chat completion model (e.g., gpt-4.1)
  • Embedding model (e.g., text-embedding-3-large)

Setup

1. Create a Python virtual environment, activate and install dependencies:

   pip install -r requirements.txt --pre
   
   # or
   # uv pip install -r requirements.txt --pre
   # if using uv (highly recommended for performance)
   # https://docs.astral.sh/uv/

2. Configure the application:

   • Copy .env.sample to .env and update it with your Azure service credentials
   • Fill in the following values:
     • Azure OpenAI endpoint and API key
     • Azure OpenAI deployment names and models
     • Azure AI Search endpoint and search key
     • Index name, knowledge source name, and knowledge agent name

3. Configure permissions (required for agentic retrieval):

   • Enable RBAC on your Azure AI Search service
   • Enable system-assigned managed identity on the Search service
   • Assign these roles to yourself:
     • Search Service Contributor
     • Search Index Data Contributor
     • Search Index Data Reader
   • Assign Cognitive Services OpenAI User role to the Search service's managed identity

Your Task

Complete the implementation in the register_agentic_search() method in agentic_rag.py:

1. Create a search index with:

   • Document fields (id, page_chunk, page_embedding_text_3_large, page_number)
   • Vector search configuration
   • Semantic search configuration

2. Upload NASA "Earth at Night" data to the index

3. Create a knowledge source that references the index

4. Create a knowledge agent that connects Azure OpenAI with the knowledge source

Running the Application

python agentic_rag.py

The application will:

1. Set up the agentic search components (once you complete the implementation)
2. Start an interactive chat session
3. Clean up resources when you exit

Project Structure

python/
├── agentic_rag.py        # Main application file
├── .env.sample           # Environment variables template (copy to .env)
├── requirements.txt      # Python dependencies
└── README.md             # This file

Tips

• Refer to the Challenge 11 challenge document for detailed implementation guidance
• Use the Azure AI Search Python SDK documentation for API reference
• The conversational interface is already implemented - focus on the retrieval components
• Test your implementation with various question types to observe agentic behavior

Resources

• Azure AI Search Python SDK
• Azure Identity Python SDK
• Agentic Retrieval Concepts