A browser-based phishing detector that uses image-based classification via Teachable Machine. Upload a screenshot of a website, and this tool will predict if it's Phishing or Safe.
-
Upload a screenshot containing a visible website URL.
-
A pre-trained MobileNet model (via Teachable Machine) classifies it as:
- ✅ Safe
- 🚫 Phishing
-
Results are displayed instantly on the webpage.
Note: Accuracy depends on the training dataset. This version is a proof-of-concept.
- HTML, CSS, JavaScript
- TensorFlow.js
- Teachable Machine (MobileNet)
- GitHub Pages for deployment
├── index.html # UI code
├── model/ # Exported TM model (model.json, metadata.json, weights.bin)
├── docs/ # 📄 Documentation & screenshots (to be added)
├── app/ # (optional) alternate versioning space for future refactor
└── README.md # You're here
| File | Description |
|---|---|
screenshots/ |
UI and sample result images |
model-info.md |
Notes on training params and classes |
walkthrough.md |
Simple usage guide with visuals |
📝 To be added by contributor
- Go to the Live Demo
- Upload a screenshot image
- See the prediction result below the image
📍 (Place these in docs/screenshots/)
- Safe site prediction ✅
- Phishing site prediction 🚫
This version uses a small dataset with basic image classification.
| Model Type | Source | Notes |
|---|---|---|
| MobileNet | Teachable Machine | Lightweight, fast, less accurate for fine-grained tasks |
- May misclassify clever phishing designs
- Cannot analyze URL text deeply
- Only works with visible screenshots
A smarter hybrid phishing detector with:
- ✅ OCR (Tesseract.js) to extract URLs
- ✅ GPT/OpenRouter AI check
- ✅ VirusTotal API verification
Follow progress in the upcoming
smart-phishing-detectorrepo.
100Psycho007 – Final year BSc Botany + Zoology student interested in AI x Cybersecurity.