AI-Powered Continuous Security
S4E Research
Here we share our product R&D, prototypes, and real-world results. Our vision is simple: faster detection, less noise, and security that runs automatically, while helping make the internet more secure.
Projects
Discover what we’re building and join us in making the internet safer.
September 2025
The Login Page Detection (LPD) system leverages machine learning to automatically identify authentication interfaces across modern, multilingual, and JavaScript-driven websites. Traditional keyword-based heuristics (e.g., “login”, “signin”, “password”) fail in dynamic or obfuscated environments, while LPD overcomes these limitations through DOM structure analysis and statistical feature extraction.
Using a manually labeled dataset of 2,791 web pages and 58 structural features, the project evaluated multiple ML models — Logistic Regression, SVM, KNN, Decision Tree, Gradient Boosting, and Random Forest — achieving 90.56% accuracy with Random Forest after hyperparameter tuning. LPD provides a scalable, language-agnostic foundation for threat intelligence, credential leak validation, and phishing detection.
September 2025
This project introduces an intelligent web crawler enhanced with machine learning to identify security vulnerabilities across websites and applications. By combining automated crawling with feature extraction and binary classification, it highlights potentially vulnerable pages for security analysts.
The crawler’s modular design supports DOM parsing, script analysis, and endpoint discovery, enabling scalable vulnerability scanning while reducing manual triage workload.
February 2026
The File Upload Detector (FUD) project introduces a hybrid, behavioral-driven approach to automatically identify file upload functionalities across the modern web. Traditional detection mechanisms rely on static keyword-based heuristics which fail on JavaScript-heavy applications with non-standard labels or dynamic form logic.
FUD overcomes these barriers by integrating Hybrid Execution Analysis with a high-performance Headless Browser Engine and intelligent DOM Injection, achieving 96.76% accuracy in detecting upload interfaces hidden within Shadow DOMs, nested iframes, or third-party dynamic libraries.
July 2025
The Next CVE Forecast system offers a data-driven way to forecast and rank technology vulnerability risks using historical CVE data. With ARIMA modeling, it predicts zero-day and known vulnerabilities, ranking technologies by projected risk to help security teams prioritize resources and address threats early.
Tested against models like LSTM and transformers, ARIMA (0,1,0) proved most accurate and efficient, achieving an 87.8% pair agreement in risk ordering. This gives teams a practical, quantitative framework for proactive vulnerability management.
July 2025
Protect yourself from phishing attacks with PhiShark, an advanced AI-powered URL scanner designed to detect and block phishing threats in real time. Leveraging cutting-edge artificial intelligence, PhiShark analyzes URLs instantly to identify malicious links before they can harm you.
Whether you’re browsing emails, social media, or websites, PhiShark acts as your vigilant digital guardian—scanning every link and alerting you to potential phishing scams, fraudulent sites, and other online threats. Stay safe, secure, and confident online with PhiShark’s seamless protection.
