01. Problem
QR codes are a common vector for phishing, but most scanners upload data to the cloud, compromising privacy. Users need a way to verify URLs without leaking their browsing history.
02. Solution Overview
- Built a 100% offline analysis engine using Kotlin Multiplatform
- Implemented an ensemble ML model + 25 heuristics for high accuracy
- Targeted 5 platforms (Android, iOS, Desktop, Web) with ~52% code sharing
- Achieved <5ms latency for real-time scanning
Build
Tech Stack
Kotlin MultiplatformCompose MultiplatformSwiftUIML KitSQLDelight
- • 5 Platform Targets (Android, iOS, JVM, JS, Wasm)
- • Ensemble ML Model (Logistic Regression + Gradient Boosting)
- • Heuristics Engine (25+ checks)
- • Red Team Mode for verification
Secure
- 100% Offline Analysis (Zero network calls)
- On-device ML & Heuristics
- No data collection
- Privacy by Architecture
03. Proof & Verification
Verified Claims
- >F1 Score: 87% on red team corpus
- >Performance: <5ms P99 latency
- >Privacy: Verified 0 network calls via ./judge/verify_offline.sh
- >Code Sharing: ~100% shared detection logic