Work 04 · Accessible applied AI
Making crop-disease AI useful in the field
The goal was not an impressive classifier. It was a practical path from a farmer’s photo to a useful next step under real access constraints.
The operating environment
A technically capable model would fail if the product ignored connectivity, device access, and trust.
I led the product around a crop-disease classifier trained on approximately 200,000 FAO images across seven disease classes and three crops. The design had to work for basic smartphones and inconsistent connectivity.
The whole product
Classification was only one step in the farmer’s decision.
We paired image-based identification with a low-bandwidth experience and expert escalation so uncertainty did not become a dead end. Product choices connected model output to language, access, and practical action.
The project reached hundreds of farmers and won the Microsoft Imagine Cup 2018 World Artificial Intelligence Award.
What stayed with me
Access and recourse are product requirements, not distribution details.
E-Agrovet shaped how I approach AI today: define who bears the cost of error, make uncertainty visible, and give people a meaningful next step when the system is unsure.
Reflection
AI capability matters only when the surrounding product makes it usable, understandable, and actionable.
- What I would do differently
- I would establish field telemetry and structured expert-feedback loops earlier to measure model drift and user outcomes over time.
- Framework connection
- Product Diagnosis Canvas · AI Trust Framework
Evidence boundary
Back to all transformations Public figures use conservative descriptions where historical records differ. The downloadable CV remains the authoritative professional record.