Inspiration
In Pakistan & India, 60% farmers lose crops to disease because they can't read English advisories. My uncle is a farmer, he sends me crop photos on WhatsApp. I wanted to build a voice AI that speaks in Urdu/Hindi.
What it does
KisanGPT - User uploads a crop photo + speaks in Urdu/Hindi "is me kya bemari hai?". Our AI:
- Detects disease using NVIDIA Nemotron Vision on Nebius AI Cloud
- Searches latest cure using Tavily API
- Replies back in Urdu/Hindi voice + text
How we built it
- Frontend: Next.js + Tailwind
- Backend: Nebius AI Cloud + Nebius Token Factory
- AI Model: NVIDIA Nemotron-4-340B-Instruct + Nemotron Vision
- Search: Tavily API for Best Use of Tavily prize
- Deployment: Fully hosted on Nebius AI Cloud
Challenges we ran into
- Urdu/Hindi voice-to-text accuracy
- Making Nemotron Vision work for desi crops
Accomplishments that we're proud of
Fully working demo on Nebius in under $50 credits, voice-first for low-literate farmers.
What we learned
How to deploy NVIDIA open models using Nebius Token Factory efficiently.
What's next for Kisan GPT
Add WhatsApp bot and IoT sensor for Physical AI track.
Log in or sign up for Devpost to join the conversation.