RAG System

A flexible and extensible Retrieval-Augmented Generation (RAG) system that supports multiple popular embedding providers, caching, adapters for various data types, and seamless integration with vector databases (ChromaDB).

🚀 Key Features

Pluggable Embedding Providers
- Supports: OpenAI, Azure OpenAI, Cohere, Google PaLM (OpenAI-compatible), SentenceTransformers, HuggingFace Transformers.
- Easily extendable to add new providers.
Flexible Data Adapters
- Easily add adapters for various data sources: JSON, PDF, Text, etc.
Vector Database Integration
- Uses ChromaDB (local or persistent).
- Automatically checks, creates, and manages collections.
Query Caching
- Supports caching of query results for speed and cost efficiency.
- Can be enabled/disabled via environment variable.
Flexible Configuration via .env
- Select provider, model, endpoint, API key, etc. through environment variables.
Easy Integration and Extension
- Clean, maintainable codebase.
- Includes examples and unit tests.

💳 License

📰 Released under the MIT license
©️ Copyright © Vic P. & Vibe Coding ❤️

🧩 Usage

💻 Installation

$ cd to/your/local/repo
$ git submodule add https://github.com/vic4key/RAG_System.git
$ pip install -r RAG_System/requirements.txt
$ python -m RAG_System.examples.rag.test # make sure it works
$ cp RAG_System/.env.example .env

⚙️ Configuration

Remember to update the .env file (see .env for all available options)

🚀 Coding

Click here to view more ...

from RAG_System import RAG_System
from RAG_System.examples.sample_faqs.sample_faqs_adapter import SampleFAQsAdapter

from dotenv import load_dotenv
load_dotenv()

similarity_threshold = 0.4

adapter = SampleFAQsAdapter()

rag = RAG_System()
rag.setup_from_adapter(
    adapter=adapter,
    adapter_params={
        "file_path": "sample_faqs.json",
    },
    collection_name="sample_faqs",
    persist_directory=f"chroma_db/sample_faqs",
    recreate=False,
)

questions = [
    "Làm sao để chỉnh sửa hoặc xóa chi tiêu?",
    "Xóa chi tiêu?",
    "How to connect my bank account?",
]

for question in questions:
    if result_items := rag.query(question):
        if best_match := result_items[0]:
            print(f"Found {len(result_items)} result items")
            print(f"Best match:")
            print(f"   Question: {best_match.metadata.get('question', '')}")
            print(f"   Answer:   {best_match.metadata.get('answer', '')}")
            print(f"   Distance: {best_match.metadata.get('distance', 1.0)} ({similarity_threshold = })")
        print(f"\nRetrieval:\n```\n{result_items}\n```")
        print(f"\nGeneration:")
        for chunk in rag.generate(question, result_items, stream=True):
            print(chunk, end="", flush=True)
    else:
        print(f"Not found")

🧩 Development

💻 Installation

$ git clone https://github.com/vic4key/RAG_System.git
$ pip install -r RAG_System/requirements.txt
$ python -m RAG_System.examples.rag.test

📚 Extending & Customization

Add a new provider:
Create a new provider class inheriting from BaseEmbeddingProvider in core/providers/ and register it in the factory.
Add a new data adapter:
Create a new adapter class inheriting from RAG_Adapter in core/adapters/ and register it in the factory.

🧪 Examples

The examples folder contains usage examples and unit tests for each data type, caching, etc.

📬 Contact

Feel free to contact via Twitter / Gmail / Blog / Website