Data

Tools

Indexes

Applications

Experiments

Awesome

An open API service indexing awesome lists of open source software.

https://github.com/tanush-g/llm-task

A privacy-focused web application that detects Personally Identifiable Information (PII) in text using advanced NLP techniques and provides AI-powered paraphrasing to help users protect their sensitive information. It leverages Named Entity Recognition (NER) and Large Language Models (LLMs) to identify and sanitize PII
https://github.com/tanush-g/llm-task

fastapi gemini named-entity-recognition ner pii-detection render spacy vercel

Last synced: 5 months ago
JSON representation

A privacy-focused web application that detects Personally Identifiable Information (PII) in text using advanced NLP techniques and provides AI-powered paraphrasing to help users protect their sensitive information. It leverages Named Entity Recognition (NER) and Large Language Models (LLMs) to identify and sanitize PII

Awesome Lists containing this project

README 

          
# PrivChat - PII Detection Web Application



A privacy-focused full-stack web application that detects Personally Identifiable Information (PII) in text using advanced NLP techniques and provides AI-powered suggestions for safer communication using Google's Gemini API.



![Demo](demo.gif)



## 🌐 Live Demo



- **Frontend:** [https://llm-task-phi.vercel.app](https://llm-task-phi.vercel.app)

- **Backend API:** [https://llm-task.onrender.com](https://llm-task.onrender.com)

- **API Documentation:** [https://llm-task.onrender.com/docs](https://llm-task.onrender.com/docs)



> **Note:** The live demo is hosted on free tiers, so it may take a few seconds to wake up after inactivity. Make sure the backend is running before testing the frontend.



## ✨ Features



- **🔍 PII Detection**: Uses spaCy's Named Entity Recognition to identify personal information including:

  - Names (PERSON)

  - Organizations (ORG)  

  - Locations (GPE)

  - Dates (DATE)

  - Money amounts (MONEY)

  - Phone numbers and other sensitive data

  

- **🤖 AI-Powered Suggestions**: Integrates with Google's Gemini 2.0 Flash model to provide:

  - Privacy-conscious text rewriting

  - Maintains original meaning while protecting sensitive data

  - Natural language output with professional tone



- **🎨 Interactive UI**: Modern, Mac-style interface featuring:

  - Real-time PII highlighting with pulsing visual effects

  - Expandable details panel showing all detected entities

  - Confidence scores for each detected PII element

  - Responsive design with smooth animations

  - Color-coded entity categories



- **⚡ Production Ready**: Fully deployed with professional infrastructure



## 🏗️ Architecture



### Backend (FastAPI)



- **Framework**: FastAPI with production-grade CORS configuration

- **NLP**: spaCy with `en_core_web_sm` model for entity recognition

- **AI Integration**: Google Gemini 2.0 Flash API for text rewriting

- **Confidence Scoring**: Custom algorithm based on entity type, length, and context

- **Deployment**: Render.com with auto-scaling and health monitoring



### Frontend (Vanilla JavaScript)



- **Design**: Mac-style window interface with sidebar navigation

- **Highlighting**: Dynamic PII highlighting with color-coded categories

- **Interactions**: Smooth animations and responsive layout

- **Accessibility**: Keyboard navigation and screen reader support

- **Deployment**: Vercel with global CDN and instant deployment



## 🚀 Quick Start (Try It Now!)



**Just visit:** [https://llm-task-phi.vercel.app](https://llm-task-phi.vercel.app)



1. Click the **"+"** button to enter input mode

2. Type or paste text containing PII (e.g., "Hi, I'm John Doe from Google in New York")

3. Click **"Send"** to analyze

4. View the AI-powered privacy-safe rewrite

5. Click **"i"** to see detailed PII analysis



## 💻 Local Development Setup



### Prerequisites



- **Python 3.8+** 

- **Google AI Studio API Key** ([Get one here](https://aistudio.google.com/apikey))



### 1. Clone the Repository



```bash

git clone https://github.com/tanush-g/llm-task.git

cd llm-task

```



### 2. Backend Setup



```bash

cd backend



# Create virtual environment (recommended)

python -m venv venv

source venv/bin/activate  # On Windows: venv\Scripts\activate



# Install Python dependencies

pip install -r requirements.txt



# Download spaCy English model

python -m spacy download en_core_web_sm



# Configure API key

cp .env.example .env

# Edit .env and add your GEMINI_API_KEY

```



### 3. Start Development Servers



```bash

# Terminal 1: Start backend

cd backend

python main.py

# Backend runs on http://localhost:8000



# Terminal 2: Serve frontend

cd frontend

python -m http.server 3000

# Frontend available at http://localhost:3000

```



### 4. Access Local Application



Navigate to `http://localhost:3000` in your web browser.



## 🎯 How to Use



1. **Input Text**: Click the "+" button to toggle input mode

2. **Analyze**: Type or paste text containing potential PII and click "Send"

3. **Review Results**: 

   - View the AI-rewritten privacy-safe version

   - See highlighted PII elements in the original text

   - Click the "i" button to view detailed PII analysis with confidence scores

4. **Privacy Protection**: Use the rewritten version for safer communication



## 📡 API Documentation



### Live API Endpoints



- **Base URL**: [https://llm-task.onrender.com](https://llm-task.onrender.com)

- **Interactive Docs**: [https://llm-task.onrender.com/docs](https://llm-task.onrender.com/docs)



### `POST /analyze`



Analyzes text for PII and generates a privacy-safe rewrite using Gemini AI.



**Request Body:**

```json

{

  "text": "Hi, I'm John Smith from Acme Corp. Call me at 555-123-4567."

}

```



**Response:**

```json

{

  "entities": [

    {

      "text": "John Smith",

      "label": "PERSON",

      "confidence": 0.95,

      "start": 8,

      "end": 18

    },

    {

      "text": "Acme Corp",

      "label": "ORG",

      "confidence": 0.85,

      "start": 24,

      "end": 33

    }

  ],

  "ai_response": "Hello! I'm a professional from a technology company. Please feel free to contact me.",

  "original_text": "Hi, I'm John Smith from Acme Corp. Call me at 555-123-4567.",

  "sanitized_text": "Hi, I'm [Name] from [Company]. Call me at [Phone]."

}

```



### `GET /health`



Health check endpoint to verify service status.



**Response:**

```json

{

  "status": "healthy",

  "spacy_loaded": true,

  "gemini_configured": true,

  "model": "gemini-2.0-flash"

}

```



## 🔧 Configuration & Deployment



### Environment Variables



```env

GEMINI_API_KEY=your_gemini_api_key_here

```



### Production Architecture



- **Backend**: Deployed on [Render.com](https://render.com) with auto-scaling

- **Frontend**: Deployed on [Vercel](https://vercel.com) with global CDN

- **Database**: None required (stateless application)

- **AI Model**: Google Gemini 2.0 Flash via API



### Model Configuration



The Gemini API is configured with optimal settings for privacy-focused text rewriting:



- **Temperature**: 0.3 (for consistent, reliable results)

- **Max tokens**: 100 (concise responses)

- **Top-p**: 0.8 (balanced creativity and coherence)



## 🛠️ Development



### Project Structure



```

llm-task/

├── backend/

│   ├── main.py              # FastAPI application with Gemini integration

│   ├── requirements.txt     # Python dependencies

│   ├── .env                 # Environment variables (API key)

│   ├── setup.sh            # Setup script for local development

│   └── test_config.py      # Configuration testing utility

├── frontend/

│   └── index.html          # Single-page application

├── demo.gif                # Demo animation

├── DEPLOYMENT_README.md    # Deployment documentation

└── README.md               # This file

```



### Adding New Entity Types



To support additional PII types, modify the entity filter in `backend/main.py`:



```python

if ent.label_ in ["PERSON", "ORG", "GPE", "DATE", "MONEY", "CARDINAL", "YOUR_NEW_TYPE"]:

```



### Customizing the UI



The frontend uses a single HTML file with embedded CSS and JavaScript:



- **Color schemes**: Modify CSS custom properties

- **Animation timing**: Adjust CSS keyframe durations  

- **Layout breakpoints**: Update responsive design rules



## 🔍 Testing



### Manual Testing



```bash

# Test health endpoint

curl https://llm-task.onrender.com/health



# Test analyze endpoint

curl -X POST "https://llm-task.onrender.com/analyze" \

     -H "Content-Type: application/json" \

     -d '{"text": "Hi, I am John Doe and I work at Google in Mountain View."}'

```



### Example API Usage



```python

import requests



# Analyze text for PII

response = requests.post(

    "https://llm-task.onrender.com/analyze",

    json={"text": "Hi, I'm Jane Smith from Microsoft in Seattle. Call me at (555) 123-4567."}

)



data = response.json()

print("Detected entities:", data["entities"])

print("AI rewrite:", data["ai_response"])

```



## 🚨 Troubleshooting



### Common Issues



#### 1. API Connection Errors

- **Issue**: Frontend can't connect to backend

- **Solution**: Check CORS configuration and ensure backend is running

- **Live Status**: Visit [https://llm-task.onrender.com/health](https://llm-task.onrender.com/health)



#### 2. Gemini API Errors

- **Issue**: "API key not configured" or quota exceeded

- **Solution**: Verify API key in environment variables or check quota limits

- **Get API Key**: [Google AI Studio](https://aistudio.google.com/apikey)



#### 3. spaCy Model Errors

- **Issue**: "spaCy model not found"

- **Local Solution**: `python -m spacy download en_core_web_sm`

- **Production**: Model is automatically installed during deployment



#### 4. Cold Start Delays

- **Issue**: First request takes 30+ seconds

- **Explanation**: Render free tier apps sleep after inactivity

- **Solution**: Upgrade to paid tier for instant responses



### Performance Notes



- **Cold Start**: First request after inactivity may take 30-60 seconds

- **Response Time**: Subsequent requests typically respond in 2-5 seconds

- **Rate Limiting**: No artificial limits, bounded by Gemini API quotas



## 🔒 Security & Privacy



### Data Handling

- **No Data Storage**: Application is stateless, no user data is persisted

- **API Security**: All API calls use HTTPS encryption

- **CORS Protection**: Configured to allow specific domains only

- **Input Validation**: All user inputs are validated and sanitized



### Privacy Considerations

- **PII Detection**: Identifies sensitive information locally using spaCy

- **AI Processing**: Text is sent to Google Gemini API for rewriting

- **No Logging**: Sensitive data is not logged or stored

- **Minimal Data**: Only necessary text is sent to external APIs



## 🚀 Deployment Details



### Backend (Render.com)

- **URL**: https://llm-task.onrender.com

- **Auto-deployment**: Triggered by GitHub commits

- **Environment**: Python 3.9+ with automatic dependency management

- **Scaling**: Auto-scales based on demand (free tier limitations apply)



### Frontend (Vercel)

- **URL**: https://llm-task-phi.vercel.app

- **CDN**: Global edge network for fast loading

- **Auto-deployment**: Triggered by GitHub commits

- **Performance**: Optimized static asset delivery



## 🤝 Contributing



We welcome contributions to improve PrivChat! Here's how to get started:



### Development Workflow



1. **Fork the repository**

2. **Create a feature branch**

   ```bash

   git checkout -b feature/amazing-feature

   ```

3. **Make your changes**

4. **Test thoroughly** (both locally and against production API)

5. **Submit a pull request**



### Areas for Contribution



- **🎨 UI/UX improvements**: Enhanced animations, better responsive design

- **🔍 PII Detection**: Support for additional entity types or languages

- **⚡ Performance**: Caching strategies, faster load times

- **🛡️ Security**: Additional validation, rate limiting

- **📚 Documentation**: Improved examples, tutorials



## 📄 License



This project is created for educational and demonstration purposes. 



## 🙏 Acknowledgments



- **[spaCy](https://spacy.io/)**: For excellent NLP capabilities and entity recognition

- **[Google Gemini](https://ai.google.dev/)**: For powerful AI text generation

- **[FastAPI](https://fastapi.tiangolo.com/)**: For the robust, high-performance backend framework

- **[Render](https://render.com/)**: For reliable backend hosting

- **[Vercel](https://vercel.com/)**: For fast frontend deployment and CDN



## 📞 Support



- **Live Demo**: [https://llm-task-phi.vercel.app](https://llm-task-phi.vercel.app)

- **API Status**: [https://llm-task.onrender.com/health](https://llm-task.onrender.com/health)

- **Issues**: [GitHub Issues](https://github.com/tanush-g/llm-task/issues)



---



**⚡ Quick Start**: Visit the [live demo](https://llm-task-phi.vercel.app) to try PrivChat immediately, or follow the [local setup guide](#-local-development-setup) to run it on your machine!