Captured from the live deployment, not a mockup
01 · The problem
Built each module around a question a farmer actually asks: what to sow, what is wrong with this leaf, when to water, how much urea. Shipped 13 modules in 12 Indian languages, free to use.
02 · How it works
- 01
Refused any crop with no district precedent by cross-checking five years of government returns, because a confident wrong sowing recommendation costs a season.
- 02
Validated against known reality: Ludhiana returns Punjab's documented wheat and rice rotation, and the onion seasonal index tracks the real cycle from 0.61 at glut to 1.67 in November.
- 03
Built for users who will not tolerate a confident wrong answer, framing 13 modules as questions a farmer actually asks and shipping in 12 Indian languages, free to use.
- 04
Put a hard guardrail in front of the model, refusing any crop the district has no five-year record of growing, because a wrong sowing recommendation costs a season.
03 · What it cost, and what it returned
Validated against ground truth, not only test metrics: Punjab's documented rotation and the real onion price cycle both return correct, alongside 93.75% disease accuracy on 10,162 images.