The officer needs a dossier. The farmer needs a sentence.
Same engine underneath. Completely different surface, because pretending both audiences need the same interface is how early-warning products fail.
Flood Ops · the officer
- Ranked cities, severity bands, rule_score to two decimals
- Evidence Mode with every term and its source
- Closest past event out of 653 labelled
- A replay they can rerun themselves
- Read on a desktop, at a desk, under pressure
KrishiOS · the household
- One verdict, in the language spoken at home
- Read aloud, because reading is not assumed
- One thing a household can actually do about it
- Shareable into a WhatsApp group in one tap
- Read on a phone, in a field, in daylight
Never a number on its own. A number without an action is just anxiety.
01 · today’s verdict
ଆଜି ବିପଦ କମ୍। ଧାନ କାଟିବା ନିରାପଦ।
Low risk today. Safe to harvest.
▶ Tap to hearଦୁଇ ଦିନ ଭିତରେ ପାଣି ବଢ଼ିପାରେ।
Water may rise within two days.
What to do
- Move stored grain above waist height
- Move livestock to the road embankment
- Charge the phone tonight
Six languages, because a warning in the wrong one is not a warning.
Odia is the anchor, not the ceiling. Migrant labour, tribal belts and cross-border basins mean a single-language product would miss the households most exposed and least served. Sadri is voice-only on purpose. Text would not reach the people who speak it.
ଓଡ଼ିଆ
Odia · text + voice
हिन्दी
Hindi · text + voice
বাংলা
Bengali · text + voice
తెలుగు
Telugu · text + voice
Sadri
Sadri · voice only, by design
English
English · admin and officer view
- Live weather and risk from the production API
- Yes
- Six-language delivery, voice included
- Yes
- Agentic orchestration behind the advice
- Yes
- Sensor card inside the app
- Demo mode
- WhatsApp delivery
- Sandbox
- Farmers enrolled through a district programme
- Not yet
A live MVP, not a deployed network.
The app works and the engine behind it is the production one. What does not exist yet is distribution: no district programme has enrolled farmers, and WhatsApp will not reach anyone until a recipient joins the sandbox.
That is a partnerships problem, not an engineering one, and it is why the first pre-seed hire after an ML engineer is a partnerships lead.