ContrastGuard scans a single APK or your entire installed app list for malware indicators, entirely on your device. No cloud calls, no telemetry, nothing ever leaves your phone.

How it works: a PyTorch Mobile neural network (trained on 120,000 labeled Android apps) runs on-device and is cross-checked against 250+ hand-written rules (dangerous permission combos, suspicious API calls, known-safe publisher whitelist) to cut down false positives. Real-time protection monitors newly installed apps in the background and notifies you of threats. Full scan history is stored locally — nothing is ever uploaded.

Tech: Kotlin, Coroutines, Room, Foreground Services, PyTorch Mobile Lite for on-device inference. No backend, no external database.

Status: active development, not production-ready. This is a personal/research project, not a certified security product and not independently audited — do not rely on it as your sole defense against malware. Known limitations include approximate (not exact) feature extraction which can cause false positives on large legitimate apps, no known-malware hash database, and static analysis only (no runtime behavior monitoring). Background reliability also varies by device manufacturer.

Purchase gets you the downloadable APK. Requires Android minSdk 26+.