Product Case Study · Agritech & Edge AI
95%-accurate crop disease detection — offline, in 10 languages.
From a Kalpita hackathon build to Kalpita KrishiAI: a fully developed, offline-first AI farming assistant that diagnoses crop disease on the device and advises every farmer in their own language.
95%+
on-device detection accuracy
10
Indian languages, voice-first
38
crop diseases across 12 crops
0
connectivity required
Overview
Six things KrishiAI puts in a farmer’s pocket
An agronomist in every pocket — offline.
KrishiAI runs crop-disease detection on the phone itself, so diagnosis, treatment, and advisory work with no signal.
95%+ detection accuracy on-device.
A MobileNetV2 ONNX model classifies 40+ disease classes from a single photo, with a confidence score on every result.
10 Indian languages, voice-first.
Speech input and spoken answers reach farmers regardless of literacy or language.
A whole season in one app.
Disease detection, hyperlocal weather, AI irrigation, mandi prices, and farm finance — unified.
Built for low-end phones.
Engineered for the smallholders that industry data places at ~80% of India’s farm workforce — roughly 200 million people.
A Kalpita hackathon-to-product story.
One more offering graduating from Kalpita’s growing portfolio of AI products and solutions.
Summary
Expert crop advice that keeps working when the network does not.
Kalpita KrishiAI puts expert crop advice in a farmer’s pocket — and keeps it working when the network does not. The product detects crop disease from one photo, on the device, at 95%+ accuracy, then explains the treatment in the farmer’s own language by voice. Built during a Kalpita hackathon and developed into a complete platform, KrishiAI folds disease detection, hyperlocal weather, irrigation advice, market prices, and farm finance into a single offline-first app — engineered for the 200 million smallholders that industry data places at the center of Indian agriculture.
Product profile
At a glance
Kalpita KrishiAI is a Kalpita innovation, built for the people most Agritech has left offline: small and medium farmers working 1–50 hectares. Industry data puts this group at roughly 80% of India’s agricultural workforce — about 200 million people — yet most digital farming tools ship in English and assume a steady connection. KrishiAI was designed for the opposite reality.
- Product
- Kalpita KrishiAI
- Industry
- Agriculture / Agritech — smallholder and cooperative farming across India.
- Size & reach
- Built for individual farmers, farmer-producer organizations (FPOs), cooperatives, and government extension programs; pilots scoped at 10,000–100,000 farmers per region.
- Core challenge
- Crop losses of 15–40% from late or inaccurate disease diagnosis, compounded by language barriers, weak connectivity, and low farm-finance visibility.
- Origin
- Built at a Kalpita hackathon, developed into a full product
Tools & technologies
The stack behind the app
- App framework
- .NET MAUI (cross-platform) · MVVM · dependency injection · service layers & repositories
- Edge AI
- MobileNetV2 ONNX via ONNX Runtime · SkiaSharp image preprocessing · on-device confidence scoring
- On-device data
- SQLite (sqlite-net-pcl) — detection history, advisories, preferences
- Cloud (optional)
- Azure SQL (sync) · Azure Speech Services · Azure OpenAI
- Architecture
- Layered, interface-based, service-oriented — extensible to new crops, languages and advisories
The challenge
The problem we solved
Indian farmers lose 15–40% of a crop to disease, and the loss usually traces to one delay: the gap between spotting a symptom and getting an expert answer.
Indian farmers lose 15–40% of a crop to disease, and the loss usually traces to one delay: the gap between spotting a symptom and getting an expert answer. KrishiAI had to close that gap under hard real-world constraints.
Scarce
rural and remote farms have little access to agronomists, so treatment decisions arrive late — after the damage.
Offline
patchy rural networks rule out any tool that needs the cloud for its core function.
Language
farmers want advice in their own language, not a foreign one.
Literacy
the app must work for first-time smartphone users with minimal onboarding.
Low-end
detection has to run accurately on low-end Android phones — limited memory, modest cameras, small batteries.
The solution
Edge AI on the phone, cloud only as an enhancement.
Kalpita built KrishiAI as an offline-first app on .NET MAUI, with the AI running on the phone itself. A layered MVVM design, interface-based services, and dependency injection keep the platform maintainable and ready to extend.
On-device disease detection
A MobileNetV2 ONNX model classifies 40+ disease classes from a photo, with SkiaSharp preprocessing and a confidence score on every prediction — no internet required.
Treatment a farmer can act on
A local knowledge base covers 38 crop diseases across 12 crop types, with organic and chemical options, dosages, severity, and prevention.
Voice-first, 10 Indian languages
Speech-to-text and text-to-speech let farmers ask and hear advice in English, Hindi, Marathi, Tamil, Telugu, Punjabi, Gujarati, Bengali, Kannada, or Malayalam.
One app for the whole season
Hyperlocal weather, AI irrigation scheduling, mandi price intelligence, and farm finance tracking sit alongside detection.
Offline by architecture
SQLite stores history and advisories on the device; Azure SQL syncs opportunistically when a connection appears.
Cloud as enhancement
Azure Speech Services and Azure OpenAI add richer conversation when online — never as a prerequisite for the core features.
Inside the product
What ships in KrishiAI
Detection is the entry point. The app carries a farmer through the whole season — weather, irrigation, market timing and money.
Edge-AI disease detection
camera or gallery image, real-time inference, confidence scoring, fully offline.
38-disease treatment library
organic and chemical remedies, dosage, affected parts, and prevention across 12 crops.
Multilingual voice assistant
ask by voice, hear the answer, in 10 Indian languages.
Hyperlocal weather intelligence
localized forecasts to plan sowing, irrigation, harvest, and crop protection.
Smart irrigation advisory
AI scheduling that reduces water waste and supports yield.
Market price intelligence
nearby mandi prices to time the sale.
Farm finance manager
income, expenses, loans, and subsidy tracking for real profit visibility.
Subsidy & scheme awareness
surfaces government benefits farmers often miss.
Inside Kalpita KrishiAI
Results
What KrishiAI delivers, measured
The product was engineered against one benchmark: expert-grade guidance on a budget phone, with the network switched off.
Detection accuracy
95%+
95%+ disease-detection accuracy computed on-device with a confidence score on every result
Connectivity
Zero required
detection, treatment, and core advisories run in airplane mode.
Languages
10
10 Indian languages by voice and text, reaching farmers regardless of literacy
Disease library
38
38 crop diseases across 12 crops covered by an actionable treatment library, extensible to more
Season coverage
One app
disease, weather, irrigation, market, and finance decisions unified in a farmer-friendly interface.
Scale
10k–100k farmers
40% crop loss by turning late diagnosis into instant, on-device detection.
Conclusion
An agronomist in every pocket — signal or not.
KrishiAI is what Kalpita’s hackathon-to-product pipeline looks like in agriculture: a hard problem, solved with edge AI, shipped to the people who need it most. By keeping detection and advice on the device and in the farmer’s language, Kalpita turns advanced AI into something a smallholder can use on a low-end phone, in a field, with no signal — on an architecture ready for more crops, more languages, and more intelligence. It is one more product graduating from Kalpita’s growing portfolio of AI products and solutions.
FAQ
Frequently asked questions
See it live
See KrishiAI detect crop disease offline.
Book a 30-minute demo and watch a single photo become a diagnosis, a treatment plan, and a spoken recommendation — with the network switched off.