Readme

This is a demo project for Retrieval-Augmented Generation (RAG) service based on LlamaIndex and FastAPI. It includes an example of how to set up and use the RAG service with a MinIO storage backend, MongoDB for metadata storage, and Qdrant for vector storage.

Prerequisites

• Docker and Docker Compose installed on your machine.
• Python 3.13 or higher installed.
• Required Python packages listed in pyproject.toml.

Setup

1. Clone the repository to your local machine.

2. Install the required Python packages using Pip:

   pip install .
   

   or if you are using uv:

    uv sync
   

3. Start the MinIO, MongoDB and Qdrant service using Docker Compose:

   cd example
   docker compose up
   

Usage

1. Modify the configuration in example/config.yaml as needed, especially the LLM settings.

   Note: we use the OpenAI-compatible models.

   Here is an example configuration for the embedding and generation models:

    rag:
        embedding_model:
            model_name: text-embedding-v4
            api_base: https://dashscope.aliyuncs.com/compatible-mode/v1
            api_key: <YOUR_API_KEY>
   
   
        generation_model:
            model_name: qwen3-next-80b-a3b-instruct
            api_base: https://dashscope.aliyuncs.com/compatible-mode/v1
            api_key: <YOUR_API_KEY>
            other_parameters:
                is_chat_model: true
   

2. Start the RAG service by running the example/main.py script:

   cd example
   python main.py
   

Test

• The test.py file contains simple tests for the RAG service functionalities. You can run it:

   python test.py
  

• The test_rag_service.py file contains unit tests for the RAG service. You can run it using pytest:

   python test_rag_service.py