SafeRoute uses real-time crowd-sourced data and on-device machine learning to provide intelligent route recommendations that help users navigate safely through unfamiliar areas. Trusted by 1.2 million users across 28 countries and featured by Apple as “App of the Day” in multiple markets.
The Challenge
Building a navigation app that processes real-time safety data from millions of users, runs ML inference on-device without draining the battery, and delivers sub-second route recalculations — all while maintaining a deceptively simple user interface that anyone could use in a moment of stress.
Our Solution
Native iOS (Swift) and Android (Kotlin) apps, each with a custom TensorFlow Lite model compiled for on-device inference. The model, trained on 800,000+ anonymised incident reports, predicts route safety scores in real time without sending location data to any server. The backend Python pipeline handles crowd-sourced data ingestion, model retraining, and OTA model updates — completely transparent to the end user.
Key Technical Achievements
- On-device ML inference in under 12ms with no battery impact
- Real-time data pipeline processing 800K+ location events per day
- Apple’s App of the Day recognition in 28 countries
- 47,000+ safety incidents prevented, per app analytics
