RLPT Framework Prototype

This project is a full-stack implementation of the RLPT (RAG Local Placeholder Tagging) framework, consisting of a Flask backend and React frontend. The system demonstrates how to protect sensitive data in Retrieval Augmented Generation (RAG) workflows by identifying PII using a local LLM, sanitizing inputs by replacing PII with placeholders, and recovering the original PII in the final output.

Project Structure

flask_rlpt/
├── backend/                 # Flask backend application
│   ├── app.py              # Main Flask application
│   ├── config.py           # Configuration loader
│   ├── requirements.txt    # Python dependencies
│   ├── data/              # Sample data for vector DB
│   ├── vector_store/      # ChromaDB persistence directory
│   ├── model/             # Model directory
│   ├── rlpt/              # Core RLPT logic
│   └── README.md          # Backend documentation
│
├── frontend/               # React frontend application
│   ├── src/               # Source code
│   ├── public/            # Static assets
│   ├── package.json       # Node.js dependencies
│   └── README.md          # Frontend documentation
│
└── README.md              # This file

Prerequisites

Backend Requirements

• Python 3.8+
• Ollama installed and running
  • Create a model from the backend/model directory
  • Run the model with Ollama
• OpenAI API key

Frontend Requirements

• Node.js 16+ and npm/yarn
• Modern web browser

Setup Instructions

1. Clone the Repository

git clone <repository-url>
cd flask_rlpt

2. Backend Setup

1. Create and Activate Virtual Environment:

   cd backend
   python -m venv venv
   # On Windows:
   venv\Scripts\activate
   # On Unix/MacOS:
   source venv/bin/activate

2. Install Python Dependencies:

   pip install -r requirements.txt

3. Configure Environment Variables: Create a .env file in the backend directory:

   OPENAI_API_KEY="your_openai_api_key_here"
   OLLAMA_BASE_URL="http://localhost:11434"
   LOCAL_NER_MODEL_NAME="DebertaV3_For_NER"
   RAG_EMBEDDING_MODEL_NAME="text-embedding-3-small"
   RAG_GENERATIVE_MODEL_NAME="gpt-4.1-mini-2025-04-14"
   CHROMA_DB_PATH="./vector_store"
   PLACEHOLDER_MAPPING_FILE="./placeholder_mapping.json"
   SAMPLE_DOCUMENTS_PATH="./data/sample_documents.txt"

4. Initialize Placeholder Mapping: Ensure backend/placeholder_mapping.json exists with initial structure:

   {
       "PERSON_NAME": {},
       "EMAIL": {},
       "PHONE_NUMBER": {},
       "ADDRESS": {},
       "URL_PERSONAL": {},
       "ID_NUM": {},
       "USERNAME": {}
   }

5. Ingest Sample Data:

   python ingest_data.py

3. Frontend Setup

1. Install Node.js Dependencies:

   cd ../frontend
   npm install
   # or if using yarn:
   yarn install

2. Configure Environment Variables: Create a .env file in the frontend directory:

   VITE_API_URL=http://localhost:5001

Running the Application

1. Start the Backend Server

# From the backend directory
python app.py

The backend server will start on http://localhost:5001

2. Start the Frontend Development Server

# From the frontend directory
npm run dev
# or if using yarn:
yarn dev

The frontend development server will start on http://localhost:5173

API Endpoints

The backend exposes the following endpoints:

• GET /health - Health check
• POST /sanitize - Sanitize input text
• POST /recover - Recover PII from text
• POST /query - Query the RAG system with RLPT protection
• POST /embed_text - Add new text to vector database

For detailed API documentation, refer to the backend README.md.

Development

Backend Development

• The backend uses Flask for the API server
• Core RLPT logic is in the backend/rlpt/ directory
• Vector store uses ChromaDB
• Local LLM (Ollama) handles PII detection

Frontend Development

• Built with React + Vite
• Uses modern React features and hooks
• ESLint configured for code quality
• Hot Module Replacement (HMR) enabled

Notes

• The backend must be running for the frontend to function properly
• Ensure Ollama is running with the correct model before starting the backend
• The system uses placeholder mapping to protect PII throughout the RAG pipeline
• Both frontend and backend have their own detailed README files for specific documentation

Troubleshooting

1. Backend Issues:

   • Check if Ollama is running and accessible
   • Verify OpenAI API key is valid
   • Ensure all environment variables are set correctly
   • Check vector store permissions

2. Frontend Issues:

   • Clear browser cache if UI is not updating
   • Check browser console for errors
   • Verify backend API URL in frontend environment variables
   • Ensure all dependencies are installed