GramSetu AI (ग्रामसेतु)

Voice-first multimodal civic & agritech copilot using Gemini 2.5 Flash to decode welfare, stop scams, and empower rural Bharat through Grandma Theory UX.

Comment

Inspiration

What it does

How we built it

Challenges we ran into

Accomplishments that we're p### 💡 Inspiration

In rural regions like Fatehpur (Uttar Pradesh) and semi-urban belts across Northern India, digital empowerment faces a severe last-mile barrier. Millions of eligible citizens miss out on vital government welfare entitlements (such as PM-Kisan, Lakhpati Didi) due to complex legalistic jargon and predatory middlemen (dalals). Furthermore, vulnerable rural households are increasingly targeted by fraudulent electricity cutoff SMS scams and predatory loan apps. We built GramSetu AI to bridge this divide with a voice-first, multimodal copilot designed around "Grandma Theory" UX—making civic intelligence intuitive for any elder or low-literacy citizen.

⚡ What it does

GramSetu AI acts as an autonomous civic, financial, and agritech assistant:

  • 📜 Multimodal Scheme Decoder (योजना मित्र): Uses Gemini 2.5 Flash to scan government notices/forms, isolating exact cash benefit amounts, required e-KYC/document checklists, and direct office locations with zero-broker dependency.
  • 🛡️ Cyber Scam Sentinel (साइबर सुरक्षा): Analyzes suspicious WhatsApp/SMS messages (e.g., fake electricity disconnection notices) with an instant 0–100% Risk Barometer and actionable vernacular safety advice.
  • 🌾 Mandi & Crop Pathology (फसल डॉक्टर): Provides visual leaf disease diagnosis, affordable organic bio-remedies, and live wholesale Mandi price benchmarks.
  • 📞 Zero-Internet USSD / IVR Simulator (*99#): Simulates telephony triage for feature-phone users without internet access.
  • 🔊 Vernacular Voice Synthesis: Low-latency Hindi and English voice readouts powered by the native Web Speech API.

🛠️ How we built it

  • Frontend & Architecture: Next.js 14 App Router, React 19, TypeScript, and Tailwind CSS for high-contrast accessibility.
  • AI Reasoning: Google Gemini 2.5 Flash API with structured JSON output schemas, fortified with client-side API key configuration and deterministic offline fallback simulation.
  • Audio & Voice: Native Web Speech API (SpeechSynthesis) and Web Audio API for zero-latency telephony tone simulation.

🚧 Challenges we ran into

  • Ensuring zero-latency audio output in regional accents without relying on heavy cloud-based TTS models.
  • Designing a "Grandma Theory" UX that presents complex welfare entitlement logic in clean, unambiguous visual cards.
  • Implementing deterministic offline clinical fallbacks so the app remains 100% operational even in low-connectivity rural zones.

🏆 Accomplishments that we're proud of

  • Achieving a 0-error TypeScript build with full WCAG AAA contrast compliance.
  • Fully functional multimodal vision and threat analysis pipeline in both Hindi (Devanagari) and English.
  • Seamless integration of the *99# IVR telephony simulator.

📚 What we learned

  • Designing for rural low-literacy users requires stripping away abstract iconography and prioritizing direct voice, color-coded risk indicators, and concrete rupee-value highlights.

🔮 What's next for GramSetu AI

  • WhatsApp Business API webhook integration for direct picture-message processing.
  • Live USSD gateway connectivity for true GSM network deployment across rural India.roud of

What we learned

What's next for GramSetu AI (ग्रामसेतु)

Built With

  • civic-tech
  • gemini-2.5-flash
  • lucide-react
  • next.js
  • react
  • tailwindcss
  • typescript
  • web-speech-api
Share this project:

Updates

Submission history