phishHunter

PhishHunter — Real-time protection from phishing scams, everywhere you browse.

Comment

Inspiration

Every day, millions of people open links without thinking twice—whether from emails, WhatsApp forwards, job offers, or random websites. Most users cannot differentiate between a real login page and a fake one, and attackers exploit this by creating convincing clones of Google, Microsoft, banking portals, and even government sites.

People unknowingly enter their passwords, OTPs, and financial details… and within seconds, hackers gain full access.

We built PhishHunter because we strongly believe cybersecurity should not depend on how “tech-savvy” someone is. The browser should protect the user—not the other way around.

What it does

PhishHunter is a demo prototype of a real-time phishing detection system. Our current version showcases:

How a browser extension can scan a webpage instantly

How we analyze the URL, page content, DOM structure, and form behavior

How an AI-powered backend can generate a risk score

How users get a clear warning overlay if a page looks suspicious

The prototype demonstrates the workflow, UI, and core detection logic we plan to expand with advanced models.

Even in demo form, it already detects several common phishing indicators.

How we built it

Our hackathon prototype consists of two components:

  1. Browser Extension (Manifest V3)

Injects a content script into every webpage

Extracts the URL + HTML content

Communicates with the backend

Displays real-time alerts and warning overlays

Popup UI shows risk level & reasons

  1. FastAPI Backend

Currently using heuristic-based detection, which analyzes:

Suspicious URL patterns

Password fields on untrusted domains

Urgent/scam-like text

Hidden iframes

Insecure or mismatched form actions

It returns a structured risk score (0–100) along with explanations.

Our goal was to build the pipeline, not the full final AI system — and that’s exactly what we achieved.

Challenges we ran into

Building even a demo version of a phishing detector came with challenges:

Balancing detection rules so the system works without being either too strict or too lenient

CORS + extension permissions, which required careful configuration

Time limitations during the hackathon

Designing an overlay system that works on any webpage without breaking layout

Ensuring the backend responds quickly enough for a real-time feel

Since phishing is a complex problem requiring huge datasets and ML training, our challenge was to show how the final product would work, even if the full AI model isn’t implemented yet.

Accomplishments that we're proud of

Built a fully working, end-to-end phishing detection system

Developed a polished browser extension UI with overlays and real-time alerts

Created a modular AI backend that can later integrate ML/CNN models

Implemented clear interpretability: users know WHY a page is risky

Achieved smooth scanning with almost zero lag

Designed a system that can realistically scale for enterprise or student use

Most importantly, We created something that can genuinely protect people from losing their accounts and money.

What we learned

How modern phishing attacks are crafted and why users fall for them

Deep understanding of:

Manifest V3 architecture

Content scripts vs background SW

Cross-origin communication

DOM parsing and security red flags

Building a real-world cybersecurity product with both frontend and backend components

How to balance usability, false positives, and security

Importance of explainability (“why something is dangerous”) in cybersecurity tools

How to design systems that work fast, locally, and securely

What's next for phishHunter

We plan to take PhishHunter to the next level with:

  1. Machine Learning-based URL classifier

Using real phishing datasets (PhishTank, OpenPhish) to train a model that boosts accuracy.

  1. CNN-based visual similarity detection

Detect pages impersonating Google, Microsoft, PayPal, etc., even if the URL appears normal.

  1. Organization dashboard

Admin-level analytics for companies and schools.

  1. Auto-block mode

Instead of warnings, high-risk pages can be instantly blocked.

  1. Edge, Firefox, and Safari support

PhishHunter is just getting started—our goal is to make the web safer for everyone, one click at a time.

Built With

Share this project:

Updates