ESC

Projects

EchoCrypt: High-Performance Side-Channel Analysis Framework

August 2024 - March 2025

GitHub repository

  • Developed a modular acoustic side-channel attack framework integrating Vision Transformers, Swin Transformers, and large language models for keystroke recovery in noisy environments.
  • Optimized inference and evaluation pipelines to achieve 90%+ recovery accuracy across unconstrained recordings.
  • Implemented LoRA fine-tuning for LLaMA models, reducing memory requirements by approximately 90% while retaining strong contextual correction performance.
  • Built reproducible PyTorch and CUDA data augmentation and benchmarking pipelines across diverse Zoom and phone audio datasets.

Interactive Malware Feature Engineering Lab

October 2024 - November 2025

lab.ali-ayati.com

  • Architected and deployed a containerized web application with Docker, Nginx, and Gunicorn for real-time malware feature experimentation.
  • Engineered a Django backend for budget-constrained Random Forest training jobs over 33+ static PE features.
  • Designed an interactive interface that provides real-time feature-importance feedback and makes malware analysis workflows more accessible.
  • Configured production deployment components to support concurrent user sessions and iterative model training.