ScamGuard AI 🎯

Basic Details

Team Name: TechDivas

Team Members

• Member 1: N Fadeela - GEC Palakkad
• Member 2: Anupama J A - GEC Palakkad

Hosted Project Link

https://scamguader.vercel.app/

Project Description

ScamGuard AI is a real-time fraud intelligence platform that identifies scam patterns in SMS and voice calls. It uses "Explainable AI" to not just flag threats, but to teach users about the psychological manipulation tactics (vishing/smishing) being used against them in real-time.

The Problem statement

Digital fraud is skyrocketing, and scammers are using increasingly sophisticated social engineering. Most people don't recognize "Artificial Urgency" or "Authority Impersonation" until it's too late. Current filters are often "black boxes" that don't provide the user with the reasoning behind a warning.

The Solution

We built a privacy-first, local-inference tool. By using the browser's built-in Web Speech and Audio APIs, we analyze text and audio for specific weighted keywords (Financial, Urgency, Threat, Reward) without sending private data to expensive third-party Cloud AI services. This ensures zero-cost, high-speed, and private analysis.

---

Technical Details

Technologies/Components Used

For Software:

• Languages used: JavaScript (ES6+), HTML5, CSS3
• Frameworks used: Express.js (Node.js)
• Libraries used: Chart.js (Real-time Analytics), Multer (File Handling)
• APIs used: Web Speech API (Speech Recognition), Web Audio API (Audio Processing)
• Tools used: VS Code, Git

---

Features

• Turbo Audio Analysis: Processes call recordings at 1.5x speed using a silent Web Audio bridge, saving users 40% of wait time without audible noise.
• Explainable AI (XAI) Panel: Categorizes scams into "Tactics" (e.g., Fear Induction, Financial Pressure) to improve user digital literacy.
• Live Mic Detection: Real-time "listening" mode for immediate analysis of ongoing phone conversations or live speech.
• Risk Trend Dashboard: A visual timeline tracking the history of scans to help users identify if they are being targeted by a persistent campaign.
• Instant Report Generator: Creates a downloadable text-based incident report for use as evidence with banks or law enforcement.

---

Implementation

For Software:

Installation

# Clone the repository
git clone [your-repo-link]

# Install dependencies
npm install

#### Run
```bash
node server.js

Project Documentation

For Software:

Screenshots (Add at least 3)

 Add caption explaining what this shows

 Add caption explaining what this shows

 Add caption explaining what this shows

Diagrams

System Architecture:

Explain your system architecture - components, data flow, tech stack interaction

Application Workflow:

Add caption explaining your workflow

---

Additional Documentation

For Web Projects with Backend:

API Documentation

Base URL: https://api.yourproject.com

Endpoints

GET /api/endpoint

• Description: [What it does]
• Parameters:
  • param1 (string): [Description]
  • param2 (integer): [Description]
• Response:

{
  "status": "success",
  "data": {}
}

POST /api/endpoint

• Description: [What it does]
• Request Body:

{
  "field1": "value1",
  "field2": "value2"
}

• Response:

{
  "status": "success",
  "message": "Operation completed"
}

[Add more endpoints as needed...]

---

For Scripts/CLI Tools:

Command Reference

Basic Usage:

python script.py [options] [arguments]

Available Commands:

• command1 [args] - Description of what command1 does
• command2 [args] - Description of what command2 does
• command3 [args] - Description of what command3 does

Options:

• -h, --help - Show help message and exit
• -v, --verbose - Enable verbose output
• -o, --output FILE - Specify output file path
• -c, --config FILE - Specify configuration file
• --version - Show version information

Examples:

# Example 1: Basic usage
python script.py input.txt

# Example 2: With verbose output
python script.py -v input.txt

# Example 3: Specify output file
python script.py -o output.txt input.txt

# Example 4: Using configuration
python script.py -c config.json --verbose input.txt

Demo Output

Example 1: Basic Processing

Input:

This is a sample input file
with multiple lines of text
for demonstration purposes

Command:

python script.py sample.txt

Output:

Processing: sample.txt
Lines processed: 3
Characters counted: 86
Status: Success
Output saved to: output.txt

Example 2: Advanced Usage

Input:

{
  "name": "test",
  "value": 123
}

Command:

python script.py -v --format json data.json

Output:

[VERBOSE] Loading configuration...
[VERBOSE] Parsing JSON input...
[VERBOSE] Processing data...
{
  "status": "success",
  "processed": true,
  "result": {
    "name": "test",
    "value": 123,
    "timestamp": "2024-02-07T10:30:00"
  }
}
[VERBOSE] Operation completed in 0.23s

---

Project Demo

Video

[Add your demo video link here - YouTube, Google Drive, etc.]

Explain what the video demonstrates - key features, user flow, technical highlights

Additional Demos

[Add any extra demo materials/links - Live site, APK download, online demo, etc.]

---

AI Tools Used (Optional - For Transparency Bonus)

Tool Used: Gemini 3 Flash / ChatGPT

Purpose:

• Asynchronous Logic Debugging: Specifically for handling the onresult events in the Web Speech API to ensure transcripts were captured before the analysis function triggered.
• Media Optimization: Assistance in configuring the AudioContext to allow for 1.5x speed "silent scanning" of uploaded audio files.
• Data Visualization: Generating the boilerplate configuration for Chart.js to map the risk score history onto a dynamic line graph.
• CSS Glassmorphism: Refined the backdrop-filter and transparency layers to ensure a high-tech, modern look for the dashboard.

Key Prompts Used:

• "How to route audio to a GainNode with 0 volume so SpeechRecognition can 'hear' it without playing sound through speakers?"
• "Create a weighted keyword matching engine in JavaScript that assigns different scores to categories like 'Urgency' and 'Financial'."
• "Debug a JavaScript error where the probability bar fails to update its background color based on the risk percentage."
• "Set up a Node.js Express route to serve an array of past scam scores for a Chart.js timeline."

Percentage of AI-generated code: Approximately 45%

Human Contributions:

• Architecture Design: Designed the "Edge-First" system where the main computation happens in the browser to protect user privacy.
• Custom Logic Implementation: Curated the specific "Scam Keyword" database and defined the heuristic weighting for different fraud types.
• Integration & Testing: Manually connected the Microphone, File Upload, and Text Input streams to a unified Result Panel.
• UI/UX Decisions: Finalized the user journey from "Message Input" to "Explainable AI Advice" to ensure the tool is educational for non-technical users.

---

Team Contributions

• N Fadeela: Engineered the core heuristic engine for text analysis, developed the live microphone detection system, and authored the comprehensive technical documentation.
• Anupama J A: Designed the Glassmorphism UI, implemented the high-speed audio/video detection bridge, and led the initial project proposal and deployment.

---

License

This project is licensed under the MIT License - see the LICENSE file for details.

Made with ❤️ at TinkerHub