Crop Talk AI Farming Assistant

Inspiration

140M Indian farmers lose ₹2L/crop to misdiagnosed pests/diseases. Agri apps are English-only, text-heavy, and ignore local pesticide brands. Farmers need voice input → photo diagnosis → Hindi audio treatment plans on JioPhones.

What it does

Voice-first AI farming doctor that:

Listens in Hindi/Marathi/English ("मेरे टमाटर की पत्तियों पर काले धब्बे")
Sees crop photos (Early Blight detection)
Speaks step-by-step treatment: "Mancozeb 75% WP, 2g/liter, 7 दिन बाद दोबारा"
Hands-free - farmer hears full plan while working
Offline demo - works without API key

Full flow: Mic → Claude AI → Local pesticides → Hindi audio (3s end-to-end)

How we built it

4hr Hackathon Stack:
├── React + TailwindCSS (mobile-first, 320px JioPhone screens)
├── Web Speech API (hi-IN SpeechRecognition + SpeechSynthesis)
├── Claude Sonnet 4 (`claude-sonnet-4-20250514`) → JSON diagnosis
├── 50+ Indian pesticides (Mancozeb, Imidacloprid 17.8 SL dosages)
├── localStorage (history + crop/disease filters)
└── Offline mock data (tomato blight, rice stress, wheat aphids)

Flow: Hindi speech → "Early Blight, 87% confidence" → "2g Mancozeb/liter" → Hindi TTS

Challenges we ran into

Chrome-only Speech API → Built offline demo fallback
Local pesticides → Manually coded 50+ Indian brands/dosages
Mobile touch targets → 44px minimum for farmer thumbs
API key exposure → Browser-direct calls (needs proxy for prod)
Hindi TTS quality → Fine-tuned rate/pitch for natural flow

Accomplishments that we're proud of

✅ Day 1 MVP: Full voice→diagnosis→audio pipeline
✅ Offline-first: Demo works without API key
✅ Local-first: Real Indian pesticides (not generic "fungicide")
✅ Mobile-first: Perfect on 320px JioPhone screens
✅ 3s flow: Mic → AI → spoken treatment before farmer walks away
✅ 5 languages: Hindi/Marathi/English ready, Tamil/Telugu next