Graduate student who fell in love with infrastructure through gaming servers and never stopped building. Studying cybersecurity, chasing cloud and security roles, and always saying yes to projects that teach me something new.
I'm a cybersecurity graduate student at Northeastern University's Khoury College of Computer Sciences with a background in computer science from BITS Pilani, Dubai.
I'm interested in cloud security, infrastructure, and active defense. Most of my projects revolve around these areas - my current research explores honeytoken-based deception for EV charging infrastructure, and I'm always picking up new things through hackathons and side projects. I also founded and chair Northeastern University's first-ever graduate ACM chapter, based at the Arlington campus.
I got into this field through gaming - I spent years running Minecraft server infrastructure from scratch, which taught me everything from Linux and firewalls to DDoS mitigation. That's where the passion started, and it hasn't stopped since.
Designed a three-stage active defense pipeline - risk scoring, adaptive response, and deception & containment - for EV charging stations. The framework targets reconnaissance, MITM, and coordinated grid attack scenarios on ISO 15118/OCPP flows.
Developed a weighted risk-scoring algorithm and adaptive honeytoken system that routed high-risk sessions into an isolated sandbox, achieving zero false-positive sandboxing of legitimate users across 50,000 simulated sessions. Engineered as a lightweight software agent (<5% CPU overhead) deployable via firmware updates on existing EVSE controllers.
Built a privacy-first desktop assistant using a local LLM (Ollama) with automatic PII redaction, agentic web search via Linkup, and 12 tools across 5 domains - all orchestrated by an autonomous Planner → Executor → Evaluator loop.
Engineered a 3-factor authentication system combining HoneyTokens, Canvas Fingerprinting, and Graphical Passwords using Node.js and Express.js. Achieved 100% detection of unauthorized credential use and 99.1% browser identification accuracy via fingerprintJS.
Secured against shoulder-surfing with a recognition-based 4×4 randomized image grid (43,680 combinations) and automated email OTP verification for all unrecognized device logins.
Developed a modified W-Net deep learning architecture, training on the BraTS dataset (369 multi-modal MRI scans). Achieved 98.54% accuracy and a Dice score of 0.9857 in classifying tumor regions (Enhancing Tumor, Tumor Core, Whole Tumor).
Engineered a sensor fusion system for a TurtleBot, integrating intent-aware SLAM and monocular camera data to improve pathfinding accuracy by 30%. Implemented a Gumbel Social Transformer model for pedestrian trajectory prediction and optimized it to reduce computational latency by 40%.
