Work / Bharat-First / KisanMind
No. 01 · Bharat-First · Agritech
One phone call.
Four satellites.
Twenty-two languages.
किसानमाइंड · KisanMind
150 million Indian farmers make ₹45 lakh crore of decisions every year · blind to crop health, mandi prices and the weather window. KisanMind hands them satellite-grade intelligence the way they already communicate. By voice. In their language. On the phone they already own.
Act I · The Problem
The data exists.
The farmer cannot reach it.
India has the satellites. India has the mandis. India has the weather.
What India does not have, in the field, is a translator.
-
i.
Cannot see what satellites see.
Sentinel-2, Sentinel-1 SAR, MODIS Terra and NASA SMAP all watch their farm. None of them speak to them.
-
ii.
Sells at the nearest mandi, not the best one.
Net profit needs price minus transport, commission, and crop-specific spoilage. Nobody hands them that math.
-
iii.
"Rain expected" is a useless forecast.
Should they irrigate? Spray? Harvest? The forecast has to know their crop, at their growth stage, at their soil moisture.
-
iv.
60% of Indian farmers don't read English or Hindi.
Most "AI for farmers" apps stop existing the moment the language changes.
Act II · The Promise
The whole product is a phone call.
A farmer in Solan opens KisanMind, taps once, and speaks. KisanMind speaks back. In Hindi. With real numbers.
मैं सोलन में टमाटर उगा रहा हूँ।
"I'm growing tomatoes in Solan."
आपकी फ़सल की सेहत मध्यम है · Sentinel-2 NDVI 0.54। टमाटर का सबसे अच्छा भाव APMC भुंतर मंडी में ₹7,500 प्रति क्विंटल है, 251 km दूर। ट्रांसपोर्ट काट के, आपको ₹6,175 मिलेंगे।
"Crop health is moderate · NDVI 0.54. Best tomato price is at APMC Bhuntar at ₹7,500/quintal, 251km away. After transport, you take home ₹6,175/quintal."
29-30 मार्च को बारिश की संभावना है। उससे पहले पानी न दें और छिड़काव न करें।
"Rain expected on 29-30 March. Don't irrigate or spray before then."
Act III · The Product
Three moves.
One conversation.
-
Step 01
Listen. Locate. Understand.
Browser captures voice (or Twilio answers a phone call). Speech-to-Text in 22 languages, GPS pins the farm, Gemini extracts crop and growth stage from the conversation.
-
Step 02
Pull every signal. In parallel.
One async fanout · four satellites, AgMarkNet's 112 commodities, Open-Meteo's 5-day forecast, Google Maps' driving distances. Cross-validated against each other before anything is said.
-
Step 03
Speak the answer back.
Gemini 3 Flash synthesises the data into specific advice (irrigate / spray / sell / wait) in the farmer's language. Cloud TTS Wavenet voices it. Farmer can ask follow-ups.
Act IV · The Fusion
Nine real sources.
One truthful answer.
Satellites, mandis, weather, maps. We don't trust any single one. Every answer is cross-validated, with a confidence score per source.
01 · Satellite
Sentinel-2
NDVI, EVI, NDWI · 10m crop health · via Google Earth Engine.
02 · Satellite
Sentinel-1 SAR
Radar soil moisture · works through clouds (C-band VV/VH backscatter).
03 · Satellite
MODIS Terra
Land surface temperature · 1km daily · heat-stress detection.
04 · Satellite
NASA SMAP L4
Root-zone soil moisture · 0·100cm deep · 9km resolution.
05 · Market
AgMarkNet
Government mandi prices from data.gov.in · 112 commodities · 90-day history.
06 · Weather
Open-Meteo
5-day forecast + 90-day historical · GDD growth-stage estimation.
07 · Logistics
Google Maps
Real driving distances + transport cost · KVK location search.
08 · Voice
Cloud Speech & TTS
STT V2 in 22 languages · Wavenet TTS · Translation v3 fallback.
09 · Synthesis
Gemini 3 Flash
Multi-turn conversation · function calling · advisory + chat + intent.
Act V · Proof
It already works.
Live · kisanmind.dmj.one
Production deployment, full voice loop, live satellite + mandi data, accessible from any browser on a 2G connection.
Twilio · +1 260-254-7946
Twilio calls the farmer back · no international charges to the farmer. Voice-only flow. No app needed.
3-tier cache · 0.13 s in-memory
L0 satellite-cache (3,788 points · O(1) lookup) → L1 in-memory (TTL) → L2 Cloud Storage. Most answers return in under 200ms.
ET AI Hackathon 2026 · Problem 5
Built for the "Domain-Specialised AI Agents" track. Submitted with full architecture diagrams, voice flow, and architecture-validated source.
The Stack
End-to-end. Production-grade. WCAG 2.2 AAA from line one.
- Python 3.11+
- FastAPI
- Next.js 16
- Gemini 3 Flash
- Gemini Live (WS)
- Google Earth Engine
- Cloud Speech-to-Text V2
- Cloud TTS Wavenet
- Cloud Translation v3
- Twilio Voice
- Cloud Storage
- data.gov.in / AgMarkNet
- Open-Meteo
- Google Maps Platform
If 150 million farmers can use it, your customers can too.
I build voice-first, satellite-fused, multilingual AI products for the India most software forgets · and the enterprises that want to reach it.