AI Wildlife Identification with Computer Vision, Azure & Offline Mobile Apps
An AI wildlife identification platform that reads leopard rosette patterns in a single photo and identifies individual leopards, operating completely offline inside national parks.
Business Requirement
Field researchers and trackers in Sri Lanka's national parks faced imperfect photos, differing left and right flanks, and zero cellular connectivity in remote areas, with identification history previously siloed across personal files.
What Mickiesoft Delivered
- Multi-stage computer vision pipeline (animal detection, flank orientation, MobileNetV4 embeddings, and SIFT/RANSAC geometric verification)
- On-device machine learning with ONNX Runtime for Android and Core ML for iOS, supporting full offline identification
- Native mobile applications: Kotlin Jetpack Compose for Android and SwiftUI for iOS with offline map support
- Cloud backend on Microsoft Azure powered by Python FastAPI, Azure SQL Database, and automated CI/CD pipelines
- Web-based admin console for reviewing sightings, managing tracker submissions, and retraining models










