KGIndex

India --:--:-- --

All work

07 · Agri-tech · hackathon prototype

TriNetra

Satellite crop-health checks, a demo price model and soil advice in local languages, in one farmer dashboard.

When
2026
Role
Sole author
Built with
FastAPI, Next.js, PyTorch, Earth Engine, Gemini
Source
KrishG7/TriNetra-Farmers-Stack ↗

See it

Market Vision showing a forecast price, a seven-day curve and a recommendation
When to sellPick a crop, a state, a market and a date, and Market Vision draws a seven-day price outlook with a hold or sell hint. Demo model.
Soil Vision with soil inputs, a cultivable verdict and three crop recommendations with fertiliser doses
What to growEnter nitrogen, phosphorus, potassium, pH and rainfall. Impossible soil is refused by a fixed rule; otherwise an AI agronomist recommends crops and fertiliser doses and explains why, in seven languages. A real answer from 28 Oct 2026.
Credit Vision with a field on a map and an eligibility result
A loan check from spaceDrop a pin on a field. Sentinel-2 satellite imagery confirms it is farmland and measures crop health against the yield being claimed. This point, in Punjab, returned a health score of 68 out of 100.

The problem

A farmer wants to know what to grow, when to sell and whether anyone will lend money. Each answer needs a different kind of data.

Numbers, with dates

  • 4 daysfrom first commit to last, 12 commits, all minegit history, 24 to 27 Jan 2026
  • 68 / 100crop-health score for one real Punjab cropland point (NDVI 0.68); a low-NDVI field scored 0 and was marked high risklive Earth Engine run, 28 Oct 2026
  • 3 toolsin one dashboard: market outlook, soil advice and satellite credit check, in seven languageslive run, 28 Oct 2026

How it works

  1. 01 · sell

    When to sell

    Pick crop, state, market and date; get a seven-day curve and a hint.

  2. 02 · grow

    What to grow

    Soil numbers in, advice out, with a rule that refuses impossible soil before any AI is asked.

  3. 03 · borrow

    Will a bank lend

    A pin on a field becomes a satellite check of whether it is farmland and how healthy the crop is.

What I built

  • Credit Vision: Sentinel-2 imagery through Google Earth Engine checks a location is farmland, then scores crop health (NDVI and NDWI) against the yield being claimed.
  • Soil Vision: a hard rule for impossible pH first, then a Gemini model turns N, P, K, pH and rainfall into a crop, fertiliser doses and a plain explanation in seven languages.
  • Market Vision: a small PyTorch model draws a seven-day price curve with a hold or sell hint.
  • A FastAPI backend, a Next.js dashboard, and OTP sign-in whose registration flow I fixed and tested.