Agritech
Shipped, mobile-first, with voice in and voice out
KisanSaathi
An agricultural assistant for Indian farmers in 8 languages. Photograph a sick plant, ask by voice, get an answer you can act on.
Problem
Farmers have questions that need answers today, in their own language, often with a phone as the only device and a photo as the only evidence. Most agricultural information online is in English, is text-only, and is not about the crop in front of them.
Approach
KisanSaathi is mobile-first and voice-first. Farmers can speak their question and hear the answer back through text-to-speech, in any of 8 Indian languages. A photo of a plant goes to a visual diagnosis flow to identify disease, and questions that need current information get web-grounded answers rather than a stale guess.
Gemini, Perplexity, and Groq share the model work, and the React and TypeScript app is built to be usable one-handed in a field.
Architecture
1. Ask
By voice or text, in any of 8 Indian languages, on a phone.
2. Diagnose
A photo of the plant goes through the visual diagnosis flow.
3. Search
Questions that need current information get web-grounded answers.
Stack
- React
- TypeScript
- Gemini
- Perplexity
- Groq
- TTS
Outcome
A working assistant with plant disease diagnosis from photos, web-grounded search, voice input and output, crop-specific advice, and real-time market information, across 8 Indian languages.
What I'd do differently
Voice changed who could use it. The same features behind a keyboard would have excluded the people the app was built for. Language support is not a feature list item here, it is the product.