Projects
Here are some of the technical projects I've worked on, ranging from cybersecurity tools to AI/ML experiments.
Genetic Neural Network Architecture Optimization
Published peer-reviewed research in the National High School Journal of Science (NHSJS). Developed a hybrid evolutionary + Bayesian optimization framework that evolves neural network architectures using genetic algorithms and then fine-tunes them with Bayesian optimization, achieving higher validation accuracy than manual tuning, random search, or standalone BO on MNIST.
Secure File Transfer Tool
A Python-based secure file transfer tool using a custom Secure Chunked Transfer (SCA) protocol that encrypts, chunks, shuffles, and verifies files with ChaCha20-Poly1305 and SHA-256 to protect confidentiality and resist traffic analysis during network transmission.
Adversarial AI Defense
Developed an end-to-end adversarial machine learning pipeline in Python that trains a CNN on MNIST, generates adversarial examples using FGSM and PGD attacks, and implements a statistical anomaly detector to identify and mitigate adversarial inputs, illustrating core concepts in AI security and model robustness.
Log Analyzer
Created a lightweight Python-based intrusion detection framework that parses Zeek system logs, normalizes events, and applies time-windowed behavioral rules to detect SSH brute force, web scanning, and other malicious activity patterns for offline security analysis.
NetFlow Attack Sequencer
Automated a Python-based event correlation engine that ingests NetFlow network flow data to build a temporal-similarity graph and extract sequences of related events, helping reveal potential multi-stage attack chains hidden within large volumes of traffic for more insightful analysis.
Understanding the Effectiveness of Deep Learning Models for Vulnerability Detection
Replicated and evaluated a state-of-the-art deep learning vulnerability detection approach by implementing behavior graph extraction and CodeBERT-based embeddings to demonstrate that incorporating inter-function semantic relationships improves recall and overall performance on real C code datasets.