ScamShield-AI

AI-powered scam detection platform for internship and job postings — protecting job seekers from fraudulent opportunities.

GitHub Repository

---

Table of Contents

• Overview
• Features
• How It Works
• Tech Stack
• Architecture
• Installation
• API Usage
• Real Response Example
• Chrome Extension
• Current Limitations
• Roadmap
• Project Structure
• Environment Variables
• License
• Author

---

Overview

ScamShield-AI helps students, fresh graduates, and job seekers avoid fake internships and recruitment scams by analyzing:

• Job descriptions
• Recruiter information
• Compensation claims
• Communication patterns
• Historical scam data

The system combines rule-based detection, DistilBERT machine learning, Retrieval-Augmented Generation (RAG) with pgvector, recruiter trust scoring, and fraud ring detection into one explainable platform.

---

Features

Feature	Description
Rule-Based Detection	75+ scam indicators (fees, urgency, fake salaries, etc.)
DistilBERT Classifier	Fine-tuned transformer model
Hybrid Scoring	Rules and ML combined for higher reliability
RAG (pgvector)	Retrieves similar scam postings as evidence
Recruiter Trust Score	Email domain and company verification
Fraud Ring Detection	Graph-based network analysis
Explainable AI	Shows why a posting is flagged
Chrome Extension	Supports LinkedIn, Internshala, Naukri, Indeed
React Dashboard	Visual analytics and history

---

How It Works

Job Posting Input
       ↓
Rule Engine (75+ rules)
       ↓
DistilBERT ML Model
       ↓
Hybrid Scoring Engine
       ↓
RAG (Similar scam retrieval)
       ↓
Recruiter Trust + Fraud Graph
       ↓
Final Scam Assessment + Explanation

---

Tech Stack

Backend

• FastAPI
• Python 3.10+
• SQLAlchemy + PostgreSQL

Machine Learning

• DistilBERT (Hugging Face)
• Scikit-learn
• Sentence Transformers (all-MiniLM-L6-v2)

Vector Database

• Supabase + pgvector

Frontend

• React.js
• HTML/CSS/JS

Extension

• Chrome Extension API

---

Architecture

+-------------+     +--------------+     +-------------+
| React Dash  |---->|   FastAPI    |---->| PostgreSQL  |
| Chrome Ext  |<----|   Backend    |<----|  + pgvector |
+-------------+     +--------------+     +-------------+
                           |
              +------------+------------+
              v            v            v
         Rule Engine   DistilBERT     RAG Retriever
                        Model
              |            |            |
              +------------+------------+
                           v
                Hybrid Scoring Engine
                           |
                           v
                Recruiter Trust + Fraud Graph
                           |
                           v
                    Final Assessment

---

Installation

1. Clone the repository

git clone https://github.com/Ayushi0750/ScamShield-AI.git
cd ScamShield-AI

2. Backend setup

cd backend
python -m venv venv

# Windows
venv\Scripts\activate

# Mac/Linux
source venv/bin/activate

pip install -r requirements.txt

3. Environment variables

Create a .env file inside /backend:

SUPABASE_URL=your_supabase_url
SUPABASE_KEY=your_supabase_anon_key
DATABASE_URL=postgresql://user:pass@localhost:5432/scamshield
MODEL_PATH=./models/distilbert_scam_model

4. Run FastAPI server

uvicorn main:app --reload

API docs: http://localhost:8000/docs

5. Frontend setup

cd ../frontend
npm install
npm start

6. Chrome Extension (Developer Mode)

1. Open chrome://extensions
2. Enable Developer mode
3. Click Load unpacked
4. Select /extension folder

---

API Usage

Endpoint

POST /api/analyze

Request body

{
  "text": "Earn ₹50,000 weekly. Pay ₹2,000 registration fee before joining."
}

---

Real Response Example

This is an actual response from the current working system.

{
  "prediction": "SCAM",
  "confidence": 75.79,
  "risk_level": "MEDIUM",
  "agreement_status": "AGREE",
  "rule_prediction": "SCAM",
  "ml_prediction": "SCAM",
  "ml_score": 95.00,
  "rag_score": 66.90,
  "similar_cases": [
    {
      "text": "Data Entry Operator Hiring made easy for Digital Jobs BIPIO...",
      "similarity": 69.87
    },
    {
      "text": " QUICK CASH - IMMEDIATE HIRING  Hey Student!...",
      "similarity": 65.96
    }
  ],
  "scam_indicators": [
    "Urgency Manipulation — Uses urgency pressure tactics",
    "Missing Company Identity — Lacks verified company identity"
  ],
  "master_risk_analysis": {
    "final_score": 58.92,
    "final_label": "HIGH_RISK",
    "agreement_status": "AGREE",
    "breakdown": {
      "rule_score": 45,
      "email_score": 40,
      "domain_score": 40,
      "trust_score": 50,
      "trust_risk_score": 50,
      "fraud_ring_score": 0,
      "rule_override_applied": false
    },
    "details": {
      "email": {
        "email": "unknown@unverified.com",
        "reputation": "UNKNOWN",
        "risk_score": 40
      },
      "domain": {
        "reputation": "UNKNOWN",
        "risk_score": 40
      },
      "rings": [
        {
          "cluster_id": 0,
          "risk_score": 0,
          "label": "SAFE",
          "size": 4
        }
      ]
    }
  }
}

Note on risk levels: The outer risk_level shows "MEDIUM" while the detailed analysis shows "HIGH_RISK" — this reflects the actual hybrid scoring logic where multiple signals are weighted differently.

---

Chrome Extension

Supported platforms:

• LinkedIn
• Internshala
• Naukri
• Indeed

How to use:

1. Install extension in developer mode
2. Navigate to any job or internship posting
3. Click the extension icon
4. View scam risk instantly

---

Current Limitations

Transparent documentation of current system state.

Area	Status
DistilBERT accuracy	Trained but not 100% accurate
RAG retrieval quality	Working but not perfect
LLM / LangChain	Not implemented

---

Roadmap

Completed:

• Rule engine (75+ rules)
• DistilBERT training
• Hybrid scoring
• RAG with pgvector
• Recruiter trust score
• Fraud ring detection
• Chrome extension (basic)
• React dashboard

Upcoming:

• Model fine-tuning improvements
• Multi-language support

---

Project Structure

ScamShield-AI/
├── backend/
│   ├── api/              # FastAPI routes
│   ├── models/           # DistilBERT and ML models
│   ├── services/         # Rule engine, hybrid scoring
│   ├── rag/              # pgvector retrieval
│   ├── trust_scoring/    # Recruiter trust
│   ├── fraud_graph/      # Fraud ring detection
│   └── main.py
├── frontend/
│   ├── src/
│   └── package.json
├── extension/
│   ├── manifest.json
│   ├── content.js
│   └── popup.html
├── datasets/
├── training/
├── notebooks/
└── README.md

---

Environment Variables

Variable	Required	Description
SUPABASE_URL	Yes	Supabase project URL
SUPABASE_KEY	Yes	Supabase anon or public key
DATABASE_URL	Yes	PostgreSQL connection string
MODEL_PATH	No	Path to DistilBERT model

---

License

MIT License — Free for educational, research, and portfolio use.
No warranty implied. Model accuracy is not guaranteed.

---

Author

Built by Ayushi Tiwari to demonstrate AI-powered fraud detection, full-stack development, and real-time scam analysis.

If this project helped you, consider giving it a star on GitHub.