Ragify - A RAG Framework

Retrieval-Augmented Generation (RAG) service for creating knowledge-based Q&A systems. Built with Flask and LangChain.

Project Structure

ragify/
├── app.py               # Flask API (GET /, GET /health, GET /config, GET /providers, POST /ask)
├── rag_chain.py         # RAG Pipeline (embeddings + vector DB + LLM)
├── ingest.py            # Vector database creation from JSON files in ./docs
├── model_providers.py   # Module for managing different model providers (OpenAI, Ollama, etc.)
├── setup_env.sh         # Script to set up virtual environment and install dependencies
├── test_models.py       # Script to test individual models
├── test_multiple_models.sh # Script to test multiple model configurations
├── requirements.txt     # Python dependencies
├── docs/                # Source documents (JSON files organized by category)
│   ├── CATEGORY_A/
│   ├── CATEGORY_B/
│   └── ...
├── storage/             # Persistent FAISS index (auto-generated)
└── .env.example         # Environment variables

Quick Setup with Script

The easiest way to get started is using the automated setup script:

# Give execution permissions to the script (if needed)
chmod +x setup_env.sh

# Run the setup script
./setup_env.sh

The script will automatically:

1. Create a Python virtual environment in the venv/ folder
2. Install all basic dependencies
3. Offer the option to install dependencies for additional providers (Ollama, HuggingFace)
4. Create a .env file with default configuration

Quick Start (manual)

If you prefer to set up the environment manually:

# 1. Create virtual environment (recommended)
python3 -m venv venv
source venv/bin/activate

# 2. Install dependencies
pip install -r requirements.txt

# 3. Add OpenAI key (if using OpenAI as provider)
cp .env.example .env
echo 'OPENAI_API_KEY=sk‑...' >> .env

# 4. Prepare the knowledge base
python ingest.py          # Reads JSON files in ./docs and builds the FAISS index

# 5. Run the API
python app.py

Using Different Model Providers

This project supports multiple language model (LLM) and embedding providers:

Available Providers:

• OpenAI: GPT models and embeddings (requires API key)
• Ollama: Local models like Llama, Mistral (requires installing Ollama)
• HuggingFace: Local models or via API

Configuration through Environment Variables

You can configure providers in the .env file:

OPENAI_API_KEY=sk-...        # Required for using OpenAI

# Provider configuration
LLM_PROVIDER=openai          # openai, ollama, huggingface 
LLM_MODEL=gpt-4o-mini        # Specific to each provider
EMBEDDING_PROVIDER=openai    # openai, ollama, huggingface
EMBEDDING_MODEL=text-embedding-3-large

Command Line Configuration

When starting the server you can specify providers:

# Example with OpenAI
python app.py --llm-provider openai --llm-model gpt-4o-mini

# Example with Ollama (requires Ollama to be installed)
python app.py --llm-provider ollama --llm-model mistral --embedding-provider ollama --embedding-model nomic-embed-text

Installing Dependencies for Different Providers

To use providers other than OpenAI, uncomment and install the necessary dependencies in requirements.txt:

# For HuggingFace
pip install langchain-huggingface transformers torch sentence-transformers accelerate

Querying the API

The API starts at http://localhost:5000 with the following documentation:

• GET / - API documentation
• GET /health - Check service status
• GET /config - View current model configuration
• GET /providers - List available providers
• POST /ask - Make knowledge base queries

Query Example:

curl -X POST http://localhost:5000/ask \
     -H 'Content-Type: application/json' \
     -d '{"message": "What is the capital of France?"}'

Response:

{
  "answer": "The capital of France is Paris. It is the largest city in France and serves as the country's political, economic, and cultural center.",
  "sources": [
    {
      "source": "/path/to/ragify/docs/GEOGRAPHY/france.json",
      "title": "France Information",
      "url": "https://example.com/france",
      "snippet": "Title: France Information\n\nDescription: France is a country located in Western Europe. Its capital is Paris, which is known for..."
    }
  ]
}

Performance Testing

The project includes tools to test and compare different model configurations:

# Test a single configuration
python test_models.py --model-name "test_name"

# Test multiple model configurations automatically
./test_multiple_models.sh

Customization

• Add more document files in docs/:
  • JSON files following the existing structure
  • Plain text (.txt) files
  • Markdown (.md) files
  • PDF files (requires PyPDF2: pip install PyPDF2)
• Run python ingest.py after adding new documents to rebuild the index.
• You can modify the prompt in rag_chain.py to adjust how queries are processed.
• To add conversational history, you can extend the system using ConversationalRetrievalChain from LangChain.

Recreating the Index with Different Embeddings

If you want to change the embedding provider, you'll need to rebuild the FAISS index:

# Rebuild the index with a specific provider and model
python ingest.py --provider ollama --model nomic-embed-text

Main Dependencies

• Flask: Lightweight web framework
• LangChain: Framework for LLM-based applications
• FAISS: Library for similarity search and vector clustering
• OpenAI/Ollama/HuggingFace: Language model and embedding providers