Event
LA Hacks
Team
Andrew Liu, Numaan Formoli
Checkit is an offline-first plant safety assistant built at LA Hacks 2026 that helps campers and hikers identify whether a plant is edible or dangerous, with no internet required. Built by Andrew Liu and Numaan Formoli, the idea came from a personal experience: one of the builders found berries on a 14-mile hike as a kid and never knew for sure whether they were poisonous.
What it solves
In the wilderness, there is no network. Most plant identification apps require internet, store location data in the cloud, and add latency that makes real-time use impractical. Eating the wrong plant can be fatal. Checkit puts instant, private, offline plant safety directly in a hiker's pocket, with region-specific knowledge compiled before the trip so it is always ready when there is no signal.
What it does
Point the camera at any plant and Checkit identifies it in real time, classifies whether it is edible or dangerous, and gives region-specific safety guidance, all offline. Before a trip, users can converse with the app online to prepare. Checkit simultaneously compiles a region-specific plant knowledge pack for that area that the local models can access without network once the user is on the trail.
How it works
Checkit runs a four-stage on-device pipeline using ZETIC Melange: YOLO11 for real-time object detection, Apple Vision for plant classification, a custom Plant300K classifier for species identification, and Gemma 4 for reasoning about edibility. Whisper tiny handles voice input for the pre-trip planning conversation with Gemini 2.5 Flash. The entire inference pipeline runs locally on-device with no network calls required in the critical safety path. Built in SwiftUI using Xcode.