Verify before you apply.
JobShield is an AI-assisted web application designed to analyze job and internship opportunities and identify potential scam indicators before a user applies or shares sensitive information.
The system accepts multiple types of inputs such as job descriptions, URLs, recruiter details, screenshots, and PDF advertisements. It combines OCR, NLP, machine learning, rule-based detection, company verification, recruiter verification, and URL analysis to generate an explainable risk assessment.
- 📄 Analyze pasted job descriptions
- 🔗 Analyze job URLs
- 📧 Analyze recruiter email addresses
- 📱 Analyze recruiter phone numbers
- 🖼️ Upload WhatsApp/Telegram/LinkedIn screenshots
- 📑 Upload job advertisement PDFs/images
- 🔍 OCR-based text extraction
- 🧠 NLP and ML-based scam detection
⚠️ Scam keyword and pattern detection- 🏢 Company verification
- 👤 Recruiter verification
- 🌐 URL and domain analysis
- 🏷️ Scam-type classification
- 📊 Multi-signal risk scoring
- 💡 Explainable risk factors and evidence
- 💾 Analysis history and user feedback
- 📈 Admin dashboard for system statistics
Fake job and internship scams are increasingly presented through professional-looking advertisements, social media messages, emails, and recruitment platforms.
Users may encounter:
- Fake recruiters
- Fake companies
- Upfront registration fees
- Security deposits
- Fake internships
- Unrealistic salary promises
- Task-based scams
- Requests for sensitive information
- Suspicious application websites
- Recruitment impersonation
JobShield aims to provide users with an additional layer of verification by analyzing multiple signals associated with a job or internship opportunity.
The main objective of JobShield is to provide an explainable risk assessment rather than simply returning a binary "Fake" or "Real" result.
The system generates:
- Risk score
- Risk level
- Possible scam category
- Detected risk signals
- Supporting evidence
- Recommended safety actions
Note: The generated score represents a system risk assessment and is not a guarantee that an opportunity is fraudulent.
USER
│
▼
REACT FRONTEND
│
┌────────────┼────────────┐
│ │ │
Text Screenshot URL
│ │ │
│ OCR │
│ │ │
└────────────┼────────────┘
▼
FLASK BACKEND
│
▼
INPUT PREPROCESSING
│
▼
TEXT EXTRACTION
│
▼
TEXT CLEANING
│
▼
INFORMATION EXTRACTION
│
┌────────────┼────────────┐
▼ ▼ ▼
Rule Detection NLP/ML Verification
│ │ │
│ │ ┌────┴────┐
│ │ │ │
│ │ Company Recruiter
│ │ Verification Verification
│ │ │ │
└────────────┼───────┴─────────┘
▼
URL ANALYSIS
│
▼
SCAM CLASSIFICATION
│
▼
MULTI-SIGNAL RISK
ENGINE
│
▼
RISK SCORE
0–100
│
▼
EXPLAINABLE AI
│
▼
EVIDENCE
│
▼
SAFETY RECOMMENDATIONS
│
▼
FINAL REPORT
│
▼
REACT FRONTEND
│
▼
USER
The project's complete workflow follows input submission → preprocessing → extraction → detection → verification → risk scoring → explanation → final report → feedback/database.
The user submits a job or internship opportunity through one of the available input methods:
Job Description
Job URL
Recruiter Email
Recruiter Phone
Screenshot
PDF/Image
The backend determines the type of input and selects the appropriate processing method.
Text → Direct processing
Screenshot → OCR
PDF → Text extraction
URL → Web/URL extraction
Email → Email-domain extraction
Screenshots and images are processed using OCR to convert visual content into machine-readable text.
Extracted text is normalized before NLP analysis.
Operations include:
- Lowercase conversion
- Symbol removal
- Duplicate-space removal
- Tokenization
- Irrelevant-word removal
- Normalization
The system extracts important entities such as:
Company
Job Title
Salary
Location
Experience
Skills
Internship Duration
Recruiter
Email
Phone
Website
Application URL
Payment Information
The system checks for suspicious patterns such as:
Registration Fee
Security Deposit
Training Fee
Processing Fee
Pay to Apply
Guaranteed Job
Guaranteed Income
Urgent Application
Limited Seats
OTP
UPI PIN
Bank Details
The cleaned job text is analyzed using an ML model.
The initial proposed approach is:
Text
↓
TF-IDF
↓
Machine Learning Classifier
↓
Scam Probability
The project can begin with Logistic Regression and later experiment with other classifiers.
The system checks:
- Whether the claimed company exists
- Official company website
- Official careers page
- Company domain
- Advertised position
- Recruiter domain consistency
A domain mismatch is treated as a risk signal rather than automatic proof of fraud.
The system analyzes:
- Recruiter email
- Email domain
- Phone number
- Company association
- Communication platform
- Claimed designation
If a URL is provided, the system analyzes:
- HTTPS
- Domain
- URL structure
- Suspicious keywords
- URL shorteners
- Redirects
- Company/domain mismatch
- Suspicious domain patterns
The system attempts to identify the possible scam category.
Possible categories include:
- Upfront Payment Scam
- Fake Recruiter
- Fake Company
- Fake Internship
- Task Scam
- Identity-Information Scam
- Fake Work-from-Home Opportunity
- Recruitment Impersonation
- Suspicious Investment/MLM-style Opportunity
- Other/Unknown
Multiple independent signals are combined:
NLP Result
Payment Detection
Company Verification
Recruiter Verification
URL Analysis
Pattern Detection
OCR Analysis
Salary/Promise Analysis
The combined signals are processed by the Risk Engine.
The system generates a score between 0 and 100.
| Score | Risk Level |
|---|---|
| 0–25 | Low Risk |
| 26–50 | Caution |
| 51–75 | Suspicious |
| 76–100 | High Risk |
Instead of displaying only the score, the system explains why the opportunity received the assessment.
Example:
🔴 Upfront payment detected
🔴 Recruiter email does not match company domain
🔴 Unusually high salary claim
🟠 Urgency language detected
🟠 Company could not be independently verified
The system generates specific evidence supporting the detected signals.
Evidence 1 → Payment requested
Evidence 2 → Recruiter domain mismatch
Evidence 3 → Company verification failed
Evidence 4 → Unusually high compensation claim
The final report contains:
Job Information
Risk Assessment
Scam Category
Detected Signals
Evidence
Recommended Actions
Users can provide feedback on whether the assessment was useful or correct.
The feedback can be stored for future model improvement.
The frontend is responsible for user interaction and visualization.
JOBSHIELD
Verify before you apply.
[ Paste Job ]
[ Upload Screenshot ]
[ Check URL ]
[ Check Recruiter ]
Analyzing opportunity...
✓ Text Analysis
✓ OCR Analysis
✓ Company Verification
✓ Recruiter Analysis
✓ URL Analysis
✓ Scam Pattern Analysis
🔴 HIGH RISK
87 / 100
Possible Scam:
Upfront Payment Scam
Detected Risks
• Payment requested
• Recruiter mismatch
• Urgency detected
Evidence
...
Recommended Actions
...
The backend is built using Python Flask and acts as the central coordinator between the frontend and analysis modules.
POST /analyze/text
POST /analyze/screenshot
POST /analyze/url
POST /analyze/email
GET /report/<id>
POST /feedback
Receive Input
↓
Validate Request
↓
Preprocess Data
↓
Run Analysis Modules
↓
Generate Risk Signals
↓
Calculate Risk Score
↓
Generate Explanation
↓
Store Result
↓
Return JSON Response
The initial ML pipeline uses:
Dataset
↓
Text Cleaning
↓
Train/Test Split
↓
TF-IDF Vectorization
↓
ML Classifier
↓
Evaluation
↓
Saved Model
Possible classifiers:
- Logistic Regression
- Naive Bayes
- Random Forest
- SVM
The project will evaluate models using:
- Accuracy
- Precision
- Recall
- F1-score
- Confusion Matrix
- ROC-AUC
The model requires examples of both legitimate and suspicious job opportunities.
- Real internships
- Graduate positions
- Work-from-home opportunities
- Genuine company postings
- Payment scams
- Fake recruiters
- Fake internships
- Task scams
- Identity theft scams
- Fake work-from-home jobs
Possible sources include:
- Kaggle
- Hugging Face
- Google Dataset Search
- Carefully constructed synthetic examples
The initial version uses SQLite.
job_id
job_title
company
description
url
date_analyzed
risk_score
risk_category
detected_signals
model_prediction
job_id
user_feedback
timestamp
company_name
official_domain
verification_status
For deployment, the database can later be migrated to PostgreSQL.
| Component | Technology |
|---|---|
| Frontend | React |
| Styling | HTML + CSS |
| Backend | Python Flask |
| Database | SQLite |
| ML | scikit-learn |
| NLP | spaCy / NLTK |
| OCR | Tesseract |
| Image Processing | OpenCV |
| PDF Extraction | PyMuPDF |
| Web Extraction | Requests + BeautifulSoup |
| Explainability | SHAP / LIME |
| Data Processing | pandas |
| Numerical Processing | NumPy |
| API Testing | Postman |
| Model Storage | joblib |
| Version Control | Git + GitHub |
The project document specifies this overall stack.
JobShield/
│
├── frontend/
│ ├── src/
│ ├── public/
│ ├── package.json
│ └── README.md
│
├── backend/
│ ├── app.py
│ ├── requirements.txt
│ │
│ ├── routes/
│ ├── preprocessing/
│ ├── nlp/
│ ├── detection/
│ ├── verification/
│ ├── risk/
│ ├── explainability/
│ ├── models/
│ └── database/
│
├── ml/
│ ├── dataset/
│ ├── notebooks/
│ ├── train.py
│ ├── evaluate.py
│ └── saved_models/
│
├── uploads/
│
├── docs/
│ ├── architecture/
│ ├── screenshots/
│ └── diagrams/
│
├── tests/
│
├── .gitignore
├── README.md
└── LICENSE
Install:
- Python 3.11
- Node.js
- Git
- VS Code
git clone <YOUR_GITHUB_REPOSITORY_URL>
cd JobShieldcd backendCreate virtual environment:
python -m venv venvActivate on Windows:
venv\Scripts\activateInstall dependencies:
pip install -r requirements.txtRun Flask:
python app.pyBackend will run locally at:
http://127.0.0.1:5000
Open another terminal:
cd frontendInstall dependencies:
npm installStart development server:
npm run devThe backend APIs can be tested using Postman.
Example:
POST /analyze/text
Request:
{
"text": "Congratulations! You are selected. Pay ₹1499 registration fee."
}Example response:
{
"risk_score": 87,
"risk_level": "High Risk",
"scam_type": "Upfront Payment Scam",
"signals": [
"Payment request detected",
"Urgency language detected"
]
}Because users may upload screenshots containing personal information, the system should minimize unnecessary data storage.
The application should:
- Avoid storing unnecessary personal information
- Mask phone numbers where possible
- Mask email addresses where possible
- Delete uploaded screenshots when persistent storage is not required
- Never store passwords
- Never store OTPs
- Never store banking credentials
- Clearly communicate what information is stored
These privacy considerations are part of the project's design.
The system should be tested with multiple types of opportunities.
Expected:
Low Risk
Expected:
High Risk
Expected:
Suspicious / High Risk
Expected:
Risk signal
but not automatically classified as fraudulent.
The system should rely on multiple signals rather than only obvious scam keywords.
An optional admin dashboard can display:
- Total opportunities analyzed
- Suspicious opportunities
- High-risk opportunities
- Scam categories
- Model accuracy
- Precision
- Recall
- F1-score
- False positives
- False negatives
Possible charts:
Risk Distribution
Scam Categories
Model Performance
Monthly Detections
Possible platforms:
- Vercel
- Netlify
Possible platforms:
- Render
- Railway
- Fly.io
Development:
SQLite
Production:
PostgreSQL
Docker can also be used for containerization.
Each team member can work on a separate module.
Frontend
Backend
ML Model
OCR
Company Verification
Recruiter Verification
Database
Testing & Documentation
Recommended Git branches:
main
frontend
backend
ml-model
ocr
verification
- Create GitHub repository
- Set up frontend
- Set up Flask backend
- Configure SQLite
- Create API structure
- Text input
- Text preprocessing
- Scam keyword detection
- Basic risk scoring
- Result dashboard
- Collect dataset
- Clean dataset
- Train TF-IDF model
- Train classifier
- Evaluate model
- Save trained model
- Screenshot upload
- OCR
- PDF extraction
- URL extraction
- Company verification
- Recruiter verification
- Domain analysis
- URL analysis
- Risk signals
- Evidence generation
- Explainable result
- Safety recommendations
- Feedback system
- Admin dashboard
- Testing
- Security/privacy
- Deployment
- Final documentation
JobShield is an experimental AI-assisted risk assessment system.
Its results should be treated as risk indicators rather than definitive proof that a job or internship is fraudulent.
Users should independently verify opportunities through official company websites and trusted communication channels before sharing sensitive information or making payments.
Additional project documentation will be maintained in:
/docs
including:
- System architecture
- Database design
- API documentation
- ML documentation
- Flowcharts
- UML diagrams
- Testing documentation
- Screenshots
- Deployment documentation
Verify before you apply.
JobShield aims to help students and job seekers identify potential warning signs in job and internship opportunities through a combination of AI, NLP, OCR, verification, and explainable risk analysis.