Selected projects
A curated selection reflecting my cybersecurity interests, research direction, and technical execution.
LLM PackageGuard
A dependency scanner that detects hallucinated or malicious packages in AI-generated code by validating imports against PyPI, OSV, and GitHub APIs. Benchmarks hallucination behavior across GPT-4, Gemini, and Cohere, and ships as a Chrome extension that flags suspicious packages inside LLM chat interfaces.
Swiper No Swiping
A Chrome extension that detects personally identifiable information (PII) in documents before they are uploaded to the web. It flags names, email addresses, phone numbers, SSNs, and financial data for review and redaction, with all processing done locally in the browser.
Neuraflix
A recommendation system exploring intelligent content suggestions through data-driven modeling and user preference patterns. Implements collaborative filtering and content-based approaches for personalized recommendations.
Face Authentication
A biometric web login built with Flask, OpenCV, and SQLAlchemy. Users enroll their face at registration and can then sign in by face, offered alongside password login much like a second factor.
Malware Analysis Lab
A controlled lab for offensive and defensive training, covering the full chain from reconnaissance and exploitation to malware detection, forensics, and incident response, including static and dynamic analysis of custom malware samples.
Smart Attendance System
Automates classroom attendance with face recognition: a camera at the door detects faces (Viola-Jones), recognizes students using HOG features and an SVM classifier in MATLAB, and generates attendance reports and defaulter lists.
Graduate research focus
My graduate research at the NYU Osiris Lab explored the intersection of artificial intelligence and cybersecurity.
Adversarial ML & LLM Supply Chain Security
My research focused on understanding how adversarial techniques can compromise machine learning models, with particular emphasis on LLM supply chain vulnerabilities. LLM PackageGuard serves as a foundational tool for identifying when LLMs suggest non-existent dependencies — a vector that could lead to supply chain attacks if malicious actors register these hallucinated package names.