Team Name: Git Happens_04
Team Leader: Tushar Singh
Hackathon: AI for Bharat Hackathon (Powered by AWS)
Track: AI for Communities, Access & Public Impact
🚨 ALIVE DEMO: Click here(https://youtu.be/eN1YQG0yROA) to watch the full 3-minute working video of Y-Connect in action! 🚨
Currently, rural citizens face massive hurdles when trying to access government schemes. Government websites are complex, require English literacy, and often demand 4G internet and PDF downloads. This creates a severe "Digital Divide" where the people who need financial assistance the most are the least equipped to navigate the portals.
Y-Connect is an AI-powered WhatsApp agent that helps rural users find and apply for government schemes using natural language. Unlike complex web portals, Y-Connect operates entirely within WhatsApp. It requires zero new app installs, works perfectly on low-end smartphones over basic 2G/3G networks, and eliminates the literacy barrier by interacting with users exactly where they are already comfortable.
For the hackathon submission, we focused on building a bulletproof, highly scalable cloud backend capable of processing concurrent WhatsApp messages with extremely low latency.
- The Gateway: Twilio WhatsApp Business API webhook integration.
- The Orchestrator: Asynchronous Python FastAPI server.
- The Memory: Redis for connection pooling, load monitoring, and queue management.
- The Knowledge Base: Qdrant Vector Database populated with real government scheme data.
- The Brain: Amazon Bedrock (Nova Lite) acting as the RAG reasoning engine to generate highly contextual, accurate, markdown-formatted scheme guides.
- The Infrastructure: Fully containerized via Docker Compose and deployed live on an AWS EC2 instance.
With the foundational RAG pipeline stabilized, our immediate development roadmap introduces frugal engineering and full multimodal accessibility:
-
Native Language Voice & Text Output: Integrating OpenAI Whisper and AWS Polly to allow users to send voice notes in local dialects (Hindi, Odia, Bhojpuri). The AI will maintain the language state and return both a playable audio reply and a detailed text guide entirely in the user's native language.
-
Ultra-Low Latency C-Layer: A custom C-language shared library (
.so) bound viactypesthat acts as a 0.01ms "Desi Safety Filter" to block spam/unsafe queries locally before they hit the expensive cloud LLM, reducing API costs by 40%. -
Instant Document Vision (OCR): Allowing users to snap photos of Income Certificates to automatically extract eligibility data.
-
Auto-Form Generation: Generating pre-filled PDF application forms directly within the WhatsApp chat.
- Backend: Python, FastAPI, Docker, Docker Compose
- Databases: Qdrant (Vector), PostgreSQL (Relational), Redis (Queue)
- AI/LLM: Amazon Bedrock (Nova Lite v1:0)
- Cloud & DevOps: Amazon Web Services (EC2, IAM), Kiro.dev (Spec-Driven Dev)
- Integrations: Twilio (WhatsApp API)
- Clone the repository:
git clone https://github.com/yourusername/Y-Connect.git
cd Y-Connect- Configure Environment:
Create a .env file and add your AWS, Twilio, Redis, and Postgres credentials. Ensure AWS_DEFAULT_REGION="us-east-1" is set for Amazon Nova access.
- Deploy the Engine:
# Launch the microservices in detached mode
sudo docker-compose up -d --build app- Ingest Scheme Data:
# Populate the Qdrant Vector DB with scheme data
sudo docker-compose exec app python -m scripts.execute_hybrid_approach- Expose Webhook:
Link your server's public IP (e.g., http://YOUR_EC2_IP:8000/twilio) to your Twilio Sandbox settings.
Built with passion for bridging India's digital divide. Special thanks to AWS, Twilio, and the open-source community.
Made with ❤️ for Bharat