DEFENSIVE MACHINE LEARNING · BROWSER INFERENCE

PhishLens

A compact phishing-URL classifier that analyzes the structure of a URL without visiting it.

Inspect a URL

The URL is parsed locally in this page. No DNS lookup, browsing request, or external URL submission is performed.

Try an example:

What the model reads

The model uses 49 lexical and structural signals extracted from the URL string. It does not inspect the destination website.

ShapeLength, path depth, punctuation, separators and repeated characters.
HostSubdomains, IP address patterns, ports, punycode and hostname structure.
LanguageCredential-related hints such as login, verify, account and update.
DestinationTLD indicators, URL shorteners, file extensions and embedded redirects.

Main branch

A small MLP trained on standardized features, designed for reproducible offline inference.

49 features→Dense 64→Dense 32→Sigmoid

Interpret results carefully

This is an experimental research model and a defensive education demo. It does not fetch pages, validate certificates, query DNS, inspect redirects, or determine whether a website is compromised. A prediction is a risk signal, not a verdict; do not open suspicious links and do not use this demo as a replacement for enterprise security controls.