Case studies
AgriVision
Offline-first native iOS. Real-time plant disease detection on device.
Source code and numbers beyond the published accuracy/FPS figures are not shared.
Case 01
On-device detection, thermal path
Problem
Real-time plant disease detection on a native iOS device. The published work is an offline-first app. The camera path is AVFoundation at 30 FPS. Thermal throttling on-device is the constraint named in the public note.
Context
Native iOS. Swift, SwiftUI, Vision Framework, CreateML. Custom object detection models. Apple Neural Engine on the camera pipeline.
What I owned
[Engineered the offline-first native iOS application. Trained the custom object detection models.]
Interface decisions
Detection runs on the device. The camera pipeline is AVFoundation at 30 FPS. The Apple Neural Engine is used on that path to prevent thermal throttling on-device. The published detection accuracy is 92%.
What I chose not to do
[Unpublished. No other rejected approach is listed here.]
Quality
[The public figures are 92% detection accuracy and a 30 FPS camera pipeline. Further checks are not shared.]
Result
An offline-first native iOS application for real-time plant disease detection. The camera pipeline is published at 30 FPS, with the Neural Engine used to prevent thermal throttling on-device.
Next
[Not published.]