Akanksha Barot / What I shipped

Four screens, and what each one had to survive.

Two farmer-facing, two for the people running the operation behind them.

Rebuilt from the products rather than screenshotted, which is why the numbers on them are the real ones and the layouts are the ones that shipped. Every constraint on the home page shows up somewhere in these four: the glare, the two bars of signal, the one wet hand, the several scripts.

FIG. 1 Farmer advisory in Marathi. Price, rainfall, a recommendation and voice playback on one screen, because a farmer checks once and puts the phone away.

FIG. 2 Field coordination. The day arrives sequenced, with what is outstanding at each stop. Sequencing produced most of the roughly 30 percent productivity gain. The map produced far less.

FIG. 3 Internal operations view. Delivery channel mix is the number that decides the roadmap: 24 percent of advisories reaching farmers by voice is a product constraint, not a statistic.

FIG. 4 What a buyer sees after scanning one code. Every line was produced by someone doing their job. Nobody filled in a compliance form afterwards.

The farmer app, screen by screen

Seven screens a farmer actually moves through.

The advisory app as it is used: diagnose a crop, check what the mandi paid this morning, buy the input, read the forecast, switch the language.

Home. Six jobs on one screen because a farmer opens the app once, standing up, and wants the day’s answer without a menu.
Select crop. Picture first, name under it. The grid is the whole screen: no search box, because typing a crop name is the step that loses people.
Crop protection. Three steps, shown before the first one is taken, so the cost of asking is known up front. Past diagnoses sit underneath as the proof it worked.
Scan the crop. The guidance is on the capture screen, not in a help page, because the photo that cannot be read is the one the farmer already walked back from.
Leaf spot identified. The match percentage is on the screen because a farmer deciding whether to spend on a spray is owed the model’s confidence, not just its answer.
Input shop. Commerce sitting inside the advisory app, not beside it. The crop health prompt is here because the moment someone is buying a pesticide is the moment to check what the problem actually is.
Market prices. The same crop at four mandis, with the movement next to the rate. A 22 percent drop and a 65 percent rise on one screen is the reason the trend link exists.
Weather. Twelve days as a table rather than a carousel of cards. Four days of thunderstorms from the 18th is a spraying decision, and it has to be legible in one look.
Select language. Seven languages across five scripts, each written in its own. A list that says “Hindi” in Latin letters is no use to somebody who cannot read Latin letters.
Back toThe challenges these came fromOrTell me what is stuck
ContactLinkedIn