>whoami

ARYAN
SINGH

Cybersecurity & AI Innovator

Poolesville High School /// SMCS Magnet

$ cat about.txt

A high school innovator working at the intersection of cybersecurity and AI. I build threat detection systems, train deep learning models, and do academic research.

$ ls skills/

CybersecurityMachine LearningPenetration TestingEthical HackingNetworkingLinux

$ echo "quote i live by"

"Reasonable people adapt themselves to the world. Unreasonable people persist in trying to adapt the world to themselves. Therefore, all progress depends on unreasonable people."

-George Bernard Shaw

scroll

$ cat experience.log

Experience

Jul 2025 - Present

Cybersecurity Researcher

w/ Dr. Mark Stamp

  • >Leading novel research on AI analysis of binaries to improve malware detection
  • >Trained a diffusion model to generate malicious opcode sequences for data augmentation
  • >Built XGBoost classifier achieving 90% accuracy, 96% recall; 91% accuracy at high noise levels where baselines dropped to 51%
  • >Writing research paper as first author for publication
Security ResearchMalware DetectionDiffusion ModelsDeep LearningXGBoost

Sep 2025 - Present

Lead Developer

Vulcan @ Cura Personalis

  • >Building communication software for nonverbal workers through the Cura Personalis Project
  • >Leading full-stack application development with React frontend and Node.js + Python backend
ReactNode.jsPythonFull-Stack

Sep 2024 - Apr 2025

Co-founder

SoteriaSecurity

  • >Created a startup providing advanced, AI-powered threat detection systems
  • >Trained a CNN based on YOLOv8 to drive the detection software
  • >Placed top 10 internationally in the Conrad Challenge Cyber-Technology & Security division
YOLOv8CNNMachine LearningStartup

$ find ./research -type f

Research

~/research/zerodetect

ZeroDetect: Boundary-Aware Generative Modeling for Zero-Day Malware Detection

PAPER IN PROGRESS - First Author w/ Dr. Mark Stamp/2025

Training diffusion models to generate malicious opcode sequences for augmenting malware classifiers. XGBoost classifier achieves 90% accuracy and 96% recall, with 91% accuracy at high noise levels where baseline models collapse to 51%.

Diffusion ModelsCybersecurityXGBoost
ZeroDetect: Boundary-Aware Generative Modeling for Zero-Day Malware Detection

$ ls -la projects/

Projects

~/projects/autored

AutoRed

  • >AI-powered red teaming tool using an LLM capable of generating JavaScript exploits
  • >Implemented transfer learning on Kimi K2 Thinking model with ~40,000 JS malware samples
  • >Successfully exploited and defaced a dummy vulnerable website 7/10 times
  • >Exploits SQL vulnerabilities, XSS, and other common web attack vectors
LLMTransfer LearningRed TeamingJavaScript