SilverGuard

Overview

Seniors are losing their life savings to "Digital Arrest" scams where fraudsters keep them on a phone call for hours, pressuring them into transferring money through GPay or NetBanking. Regular spam blockers don't help here because the problem isn't spam, it's social engineering. SilverGuard watches for the behavior itself, not just phone numbers.

The key idea is what I call the Coercion Interlock: if someone is on a phone call and opens a banking app at the same time, that's almost always fraud in progress for elderly users. The app triggers an alert immediately, reads the screen for risky keywords, throws up a full-screen warning, and notifies family members through Gmail via N8N webhooks.

Key Features

• Coercion Interlock™: physically detects simultaneous call + banking app usage and triggers critical alert
• Real-time text analysis that reads screen context (WhatsApp/SMS) for risky keywords like "OTP", "Police", "RBI"
• Full-screen blocking overlay that interrupts the scam flow and prevents transactions
• Family Guardian: instantly sends Gmail alerts to family members via N8N cloud webhook
• AI-powered text verification using Groq LLaMA-3-70b for advanced scam content analysis
• Local scoring combined with cloud AI for fast, reliable fraud detection

How It's Built

SilverGuard uses a multi-layer architecture optimized for real-time detection:

• Mobile App: Flutter (Dart) provides the UI layer for permissions setup, alert dashboard, and family contact management
• Native Modules: Kotlin handles the heavy lifting via Android AccessibilityService (screen text reading) and TelephonyManager (call state detection). These native modules detect the coercion pattern in real time
• AI Layer: Groq API with LLaMA-3-70b analyzes extracted text for advanced scam verification when local keyword scoring is uncertain
• Automation: N8N cloud webhooks handle the alert pipeline, triggering Gmail notifications to registered family guardians when threats are detected

Interesting Challenges

• AccessibilityService Permissions: Android's AccessibilityService is powerful but heavily restricted. Getting reliable screen text extraction while respecting platform security constraints required careful Kotlin implementation
• Real-Time Detection: The coercion pattern must be detected instantly. Any delay means money could already be transferred. Optimizing the call state + app detection pipeline for sub-second alerts was critical
• False Positives: Legitimate use cases (calling someone while checking your bank balance) needed careful tuning of the interlock logic and keyword scoring thresholds to minimize disruption

Screenshots