🏛️ MySarkar - AI-Powered Government Scheme Discovery Platform

Bridging Citizens and Government Services through Intelligent Technology

---

🌟 Moto

"If AI can revolutionize e-commerce recommendations, why not help citizens discover the government schemes they deserve?"

MySarkar is a comprehensive multilingual AI platform that makes government scheme discovery intelligent, accessible, and inclusive — supporting everything from smartphone apps to keypad phone IVR calls.

---

Team AKAASHVANI - Koustav, Swetanjana, Vidhya, Tanishq, Shahid

🧩 Core Features

🔍 1. Intelligent Scheme Discovery

✅ Vector-powered search using:

• 🧠 Semantic embeddings with Xenova/all-MiniLM-L6-v2 + multilingual transformers
• 📍 Location-based filtering (state/district)
• 👤 Profile-based personalization
• 🎯 Real-time eligibility checking

{
  "query": "farmer loan scheme west bengal",
  "results": [
    {
      "name": "PM-KISAN Samman Nidhi",
      "eligibility_score": 0.94,
      "benefits": "₹6,000/year direct transfer"
    }
  ]
}

---

🤖 2. Universal Government AI Assistant

Powered by Google Gemini + Hugging Face Models with comprehensive knowledge:

• 🏥 Healthcare & Insurance (Ayushman Bharat, PMJJBY)
• 🎓 Education & Scholarships (NSP, PM YASASVI)
• 💼 Employment & Skills (MGNREGA, PMKVY)
• 💰 Financial Services (Jan Dhan, Mudra Loans)
• 📄 Digital Services (Aadhaar, PAN, Passport)

{
  "query": "West Bengal health insurance schemes",
  "response": "Available schemes: Ayushman Bharat (₹5L coverage), Swasthya Sathi (₹5L state scheme)..."
}

---

📞 3. Multilingual IVR System (Twilio + Ngrok)

Press 1 for scheme information
Press 2 for eligibility check
Press 3 to speak with AI assistant

Supports keypad phones with:

• 🔄 Twilio Voice API integration
• 🌐 Ngrok tunneling for local development
• 🗣️ Text-to-Speech with IndicTrans2 for Indian languages
• 📞 DTMF navigation for feature phones
• 🧠 Hugging Face models for voice processing

---

📄 4. Document Intelligence & OCR

Smart form assistance using:

• 👁️ Tesseract OCR for document scanning
• 🧠 AI-powered field extraction
• 📝 Step-by-step form filling guidance
• 🔊 Audio instructions via TTS

{
  "extracted_fields": ["name", "aadhaar", "income"],
  "help_text": "Fill Aadhaar as 12-digit number without spaces",
  "audio_url": "/audio/form_help_hindi.mp3"
}

---

👥 5. Community Engagement Platform

Role-based social features:

• 💬 Community posts and discussions
• 🏷️ Special badges for govt officials & NGOs
• 📊 AI sentiment analysis of community feedback
• 🚫 Hate speech detection and moderation
• 🔔 Real-time notification system

---

⚙️ Tech Stack

---

🏗️ Architecture Overview

┌─────────────────┐    ┌─────────────────┐    ┌─────────────────┐
│   React Frontend │────│  Node.js Backend │────│  FastAPI AI Hub │
│   (Port 3000)   │    │   (Port 5000)   │    │   (Port 8000)   │
└─────────────────┘    └─────────────────┘    └─────────────────┘
         │                       │                       │
         │              ┌─────────────────┐              │
         └──────────────│  MongoDB Atlas  │──────────────┘
                        │ (Vector Search) │
                        └─────────────────┘
                                 │
                    ┌─────────────────────────┐
                    │   Twilio IVR System     │
                    │   (Ngrok Tunneling)     │
                    └─────────────────────────┘

---

🚀 Quick Start

📋 Prerequisites

Node.js 18+, Python 3.9+, MongoDB Atlas account
Twilio account, Google API keys, Ngrok

🔧 Installation

1️⃣ Backend Setup

cd backend
npm install
cp .env.example .env
# Configure: MONGODB_URI, JWT_SECRET, CORS_ORIGIN
npm run dev  # Runs on port 5000

2️⃣ Frontend Setup

cd frontend
npm install
cp .env.example .env
# Configure: VITE_BACKEND_URL, VITE_GEMINI_API_KEY
npm run dev  # Runs on port 3000

3️⃣ AI Services Setup

cd fastAPI
python -m venv myenv
source myenv/bin/activate  # Windows: myenv\Scripts\activate
pip install -r requirements.txt
cp .env.example .env
# Configure: GOOGLE_API_KEY, BACKEND_URL
python -m uvicorn app.main:app --reload  # Runs on port 8000

4️⃣ IVR System Setup

# Install ngrok globally
npm install -g ngrok

# Expose FastAPI to internet
ngrok http 8000

# Configure Twilio webhook URL:
# https://your-ngrok-url.ngrok.io/ivr/voice

---

📊 API Endpoints

🔐 Authentication

🎯 Schemes

🤖 AI Services

📞 IVR System

---

🧠 AI/ML Components

🔍 Vector Search Engine

// Semantic scheme matching with state-of-the-art embeddings
const embedder = await pipeline("feature-extraction", "Xenova/all-MiniLM-L6-v2");
const queryEmbedding = await embedder(userQuery);
const results = await collection.aggregate([
  {
    $vectorSearch: {
      queryVector: queryEmbedding,
      path: "embedding",
      numCandidates: 200,
      limit: 5
    }
  }
]);

🌐 Advanced Translation Pipeline

# IndicTrans2 for Indian languages
from IndicTransToolkit import IndicProcessor
ip = IndicProcessor(inference=True)
input_sentences = ["Government scheme information"]
batch = ip.preprocess_batch(input_sentences, src_lang="eng_Latn", tgt_lang="hin_Deva")
translated = model(batch)
output = ip.postprocess_batch(translated, lang="hin_Deva")

🧠 Multi-Model AI Pipeline

# Hugging Face Transformers integration
from transformers import pipeline, AutoTokenizer, AutoModel

# Sentiment analysis for community posts
sentiment_analyzer = pipeline("sentiment-analysis", 
                            model="cardiffnlp/twitter-roberta-base-sentiment-latest")

# Multilingual embeddings
embedding_model = AutoModel.from_pretrained("sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2")

# Google Gemini for conversational AI
model = genai.GenerativeModel('gemini-1.5-flash')
response = model.generate_content(f"""
You are an Indian Government Services Assistant.
Query: {user_query}
Schemes: {relevant_schemes}
Respond directly with helpful information.
""")

👁️ Computer Vision & OCR

# Advanced document processing pipeline
import pytesseract
from PIL import Image
from transformers import pipeline

# OCR with preprocessing
ocr_result = pytesseract.image_to_string(preprocessed_image, lang='eng+hin+ben')

# Document classification
doc_classifier = pipeline("image-classification", 
                         model="microsoft/dit-base-finetuned-rvlcdip")

# Form field extraction with NER
ner_pipeline = pipeline("ner", 
                       model="dbmdz/bert-large-cased-finetuned-conll03-english")
fields = ner_pipeline(ocr_result)

🛡️ Content Moderation

# Custom hate speech detection model
import pickle
from sklearn.feature_extraction.text import TfidfVectorizer

# Load pre-trained models
tfidf = pickle.load(open('vectorizer.pkl', 'rb'))
hate_model = pickle.load(open('model.pkl', 'rb'))

# Multi-language hate speech detection
transformed_text = transform_text(user_input)
vector_input = tfidf.transform([transformed_text])
is_hate_speech = hate_model.predict(vector_input)[0]

---

🎭 User Roles & Permissions

---

📱 Multi-Platform Support

🌐 Web Application

• Responsive design for desktop/mobile
• Progressive Web App (PWA) ready
• Offline scheme browsing capability

📞 IVR System Flow

📞 User calls Twilio number
    ↓
🎵 Welcome message in preferred language
    ↓
🔢 DTMF menu options
    ↓ (Press 1)
🔍 Scheme information service
    ↓ (Press 2)  
✅ Eligibility checking
    ↓ (Press 3)
🤖 AI assistant conversation

📱 Mobile Features

• Touch-friendly interface
• Voice input support
• Camera document scanning
• Push notifications

---

🛡️ Security & Compliance

🔐 Authentication & Authorization

// JWT-based security
const token = jwt.sign({ userId, role }, JWT_SECRET, { expiresIn: '7d' });
// Role-based access control
const hasPermission = user.permissions.includes('manage_schemes');

🚫 Content Moderation

# Hate speech detection
transformed_text = transform_text(user_input)
vector_input = tfidf.transform([transformed_text])
is_hate_speech = hate_model.predict(vector_input)[0]

📊 Data Privacy

• GDPR-compliant data handling
• Encrypted sensitive information
• Audit logs for all operations
• User consent management

---

📈 Real-World Impact

---

🧪 Sample Workflows

✅ Scheme Discovery

POST /api/v1/schemes/search
{
  "query": "farmer subsidy odisha",
  "user_profile": {
    "state": "Odisha",
    "category": "farmer",
    "income": 200000
  }
}

🤖 AI Chat Assistance

POST /chat
{
  "message": "मुझे स्वास्थ्य बीमा चाहिए",
  "language": "Hindi"
}

📄 Document Analysis

POST /analyze-form
{
  "image_data": "base64_encoded_image",
  "language": "Bengali"
}

---

🔄 Development Workflow

🧪 Testing

# Backend tests
cd backend && npm test

# AI services tests
cd fastAPI && python test_services.py

# Frontend tests
cd frontend && npm run test

🚀 Deployment

# Production build
npm run build

# Docker deployment
docker-compose up -d

# Environment setup
cp .env.production .env

---

📊 Monitoring & Analytics

📈 Key Metrics

• Scheme discovery success rate
• User engagement analytics
• IVR call completion rates
• AI response accuracy scores
• Community sentiment trends

🔍 Logging

// Structured logging
logger.info('Scheme search', {
  userId,
  query,
  resultsCount,
  responseTime: Date.now() - startTime
});

---

📊 Monitoring & Logging

---

🌟 Future Roadmap

🔮 Phase 2 Features

• 🔗 Direct government API integrations
• 📱 Native mobile applications
• 🎙️ Advanced voice interactions
• 🤖 Chatbot for WhatsApp/Telegram
• 📊 Advanced analytics dashboard

🌍 Expansion Plans

• Support for all 22 official Indian languages
• Integration with state government portals
• Blockchain-based application tracking
• AI-powered policy recommendation engine

---

🤝 Contributing

🛠️ Development Setup

1. Fork the repository
2. Create feature branch: git checkout -b feature/amazing-feature
3. Commit changes: git commit -m 'Add amazing feature'
4. Push to branch: git push origin feature/amazing-feature
5. Open a Pull Request

📝 Code Standards

• ESLint for JavaScript/React
• Black formatter for Python
• Conventional commits
• 80%+ test coverage

---

📄 License

MIT © 2025 MySarkar Team
Built for Digital India Initiative

---

🙏 Acknowledgments

• Google Gemini for conversational AI
• MongoDB Atlas for vector search capabilities
• Twilio for voice communication infrastructure
• Hugging Face for transformer models
• Government of India for open data initiatives

---

🚀 Making Government Services Accessible to Every Indian Citizen