AI Software Developer & Researcher
I build intelligent systems at the intersection of machine learning and cybersecurity — creating models that are accurate, robust, and explainable enough to trust in production. Currently a Research Assistant at the University of North Texas AI for Cybersecurity Lab.
My work sits at a genuinely exciting frontier — applying deep learning, adversarial machine learning, and generative models to problems that matter. At the UNT AI for Cybersecurity Lab, I design and evaluate systems capable of detecting novel intrusions, classifying malicious traffic, and generating synthetic datasets for training.
I care deeply about the gap between research and production. A model that works in a notebook but can't be deployed, explained, or trusted under adversarial conditions isn't really finished. That engineering rigor is what I bring to every project.
Outside the lab, I'm drawn to the ethical dimensions of autonomous systems, keeping up with the ML literature, and building things that solve real problems.
Full Resume ↗A handful of things I've built. Each one is a real system, not a demo.
A fine-tuned BERT model trained on network log sequences to classify and flag adversarial intrusion patterns with high recall. The central challenge was translating raw network telemetry into token sequences a language model could reason about — and then making those predictions interpretable. I integrated attention visualization so security analysts can see exactly which events triggered a classification, not just what the verdict was.
An end-to-end ML pipeline that extracts lexical, structural, and domain-based features from raw URLs and classifies phishing attempts in real time. I built this as a production-ready system: the model is served via a Flask REST API, paired with a lightweight analyst dashboard, and deployed with CI/CD through GitHub Actions. It was designed to be fast enough for inline inspection.
One of the most persistent problems in cybersecurity ML is the scarcity of labeled training data — organizations rarely want to share real incident logs. I built a conditional GAN that generates statistically realistic synthetic network flow records, conditioned on traffic class labels. The generated datasets have been used to augment training for downstream anomaly detection models.
A developer tool that plugs into GitHub Pull Requests and uses LLM APIs to analyze code diffs for security vulnerabilities, logic errors, and style inconsistencies. Rather than surfacing noisy lint warnings, it provides contextual, plain-language explanations of issues — closer to a thoughtful colleague than a linter.
Technologies and domains I work in regularly.
Open to full-time roles, internships, and research collaborations. I read every message — don't hesitate to reach out.