plantscope
An on-device iOS app that helps urban Indian home growers diagnose plant-health issues from a photo.
overview
Home growers often notice something is wrong with a plant but don't have a fast, clear way to figure out what they're looking at. PlantScope turns that moment into a lightweight, private diagnosis flow on iPhone.
ML approach
Two models, chained. A cascaded MobileNetV3 feeds into an EfficientNet-B4, which keeps inference on-device without giving up the capacity to tell visually similar plant-health conditions apart.
Training runs on a proprietary iNaturalist dataset spanning 42 plant categories, with the product direction shaped by what urban Indian home growers deal with day to day.
why on-device
Keeping the diagnosis pipeline on the device makes the experience quicker and more private. It also makes PlantScope useful in the everyday conditions where an internet connection shouldn't decide whether an app can help.
What I'm learning
PlantScope has pushed me into dataset work, model trade-offs, and iOS implementation all at once, and the hardest part is making a technically complex system feel simple enough to use on a balcony or in a kitchen.