Demo Pictures

SilverGuard

Real-Time Financial Voice/Text Phishing Defense

Live voice detection. Automatic SMS interception. Fraud account verification. Guardian alerts.

SilverGuard is a full-stack, real-time scam defense system built specifically to protect seniors from financial fraud — before money is lost.

We don’t just detect scams. We intervene during them.

🚨 The Problem

Adults 60+ lost $3.4 billion to fraud in 2023.

Scammers exploit:

Authority impersonation (IRS, police, prosecutors)
Urgency and fear tactics
OTP and credential harvesting
Fake merchants and fraudulent bank accounts

By the time victims realize it’s a scam — the money is gone.

Existing solutions are:

Reactive
Fragmented
Too technical
Not built for seniors

🛡️ Our Solution: Real-Time, Multi-Layer Scam Defense

SilverGuard combines:

Live AI voice detection
Unified Calibrated Phishing Detection Engine (Trained AI Model)
Multi-modal Scam Analysis
Automatic Android SMS Phishing Interception
Fraud Account & Merchant Verification (Capital One Nessie API)
Guardian notification & Intervention System
Interactive Scam Awareness Training Game
AI Voicebot — Real-Time Legal Scam Guidance

All powered by a single, calibrated phishing-detection engine.

1️⃣ Live Voice Scam Detection (Real-Time AI During Calls)

SilverGuard analyzes conversations as they happen.

When the user taps the mic:

Audio is recorded in 2-second chunks
Whisper transcribes speech instantly
The accumulated transcript is continuously sent to our ML model
The phishing probability recalculates in real time
The RiskGauge updates dynamically

This is a continuous model scoring — not a one-time check.

As the scammer escalates pressure, the risk percentage rises live on screen.

2️⃣ Unified Calibrated Phishing Detection Engine

Every protection feature in SilverGuard uses the same production-ready ML model.

Model Architecture

TF-IDF + Logistic Regression
Calibrated with CalibratedClassifierCV
FastAPI deployment
Millisecond inference

Performance (Held-Out Test Set)

AUC: 0.98+
Accuracy: 97.9%
Low false positive rate
Balanced precision & recall

Probabilities are statistically calibrated — meaning an 85% score truly reflects high risk.
This makes it safe for UI warnings, guardian alerts, and future financial integrations.

3️⃣ Multi-Modal Scam Diagnosis

Scams don’t come in just one format — neither do we.

SilverGuard analyzes:

💬 Text Messages

Paste suspicious content → instant phishing score.

📷 Screenshots (GPT-4o-mini Vision OCR)

Image uploaded → OCR extracts text → phishing model evaluates it.

🎙 Voicemails (OpenAI Whisper STT)

Audio uploaded → Whisper transcription → phishing model evaluates transcript.

All inputs flow into the same calibrated risk engine.

4️⃣ Automatic Android SMS Phishing Interception

SilverGuard proactively monitors incoming SMS messages.

When a new message arrives:

It is analyzed in the background using our phishing model
If risk ≥ 70%, a high-priority alert is triggered
The notification deep-links directly into the Diagnosis screen
The result is preloaded — no re-analysis required

This transforms phishing detection from manual checking to automatic protection.

5️⃣ Fraud Scanner — Transfer Protection (Capital One Nessie API)

Even if a scammer convinces someone to send money, SilverGuard adds a final line of defense.

Before transferring funds, users can check:

Account IDs
Merchant names

Powered by Capital One’s Nessie API

SilverGuard integrates with the Nessie developer sandbox (api.nessieisreal.com) to:

Verify account IDs
Verify merchants
Cross-reference against a local fraud intelligence database
Detect impersonation attempts using fuzzy matching

Result

🔴 High Risk Detected (with fraud reason)
🟢 Verified Secure

Users can also submit new fraud reports to expand fraud intelligence.

This bridges phishing detection with financial transfer protection.

6️⃣ Guardian Intervention System

Scam protection shouldn’t rely on seniors alone.

Passwordless OTP Login

Phone number authentication
Twilio 6-digit verification
JWT-based secure session

Secure Guardian Pairing

Senior generates 6-digit time-limited code
Guardian enters code
Accounts linked bidirectionally

Real-Time Guardian Alerts

When phishing ≥ 70%:

Guardian receives alert
Opens preloaded scam analysis
Can immediately call the senior

This creates a real-world intervention layer.

7️⃣ Scam Awareness Training Game

Prevention matters.

SilverGuard includes an interactive phishing training system:

SMS-style scenarios
Email-style scenarios
Voice call transcript simulations
Immediate feedback + explanations
Performance badges
Anti-replay backend logic

Users don’t just get protected — they get educated.

8️⃣ AI Voicebot — Real-Time Legal Scam Guidance (Powered by OpenAI)

SilverGuard includes an AI Voicebot that gives seniors immediate, legally grounded phishing response guidance — by voice or text.

🔊 Whisper (Speech-to-Text)

If the user speaks, OpenAI Whisper STT converts their voice into text in real time (/voice-chat endpoint), enabling hands-free interaction.

🧠 GPT-4o-mini (RAG-Based Legal Guidance)

We use GPT-4o-mini with Retrieval-Augmented Generation (RAG).
When a question is asked, the system retrieves relevant sections from an Illinois voice phishing legal guide (section_index.json) and generates responses grounded in actual legal procedures — not generic chatbot answers.

This ensures: