{"id":2223,"date":"2026-03-21T16:22:59","date_gmt":"2026-03-21T16:22:59","guid":{"rendered":"https:\/\/sites.wp.odu.edu\/locky\/?page_id=2223"},"modified":"2026-06-10T14:22:23","modified_gmt":"2026-06-10T14:22:23","slug":"milestone-projects","status":"publish","type":"page","link":"https:\/\/sites.wp.odu.edu\/locky\/milestone-projects\/","title":{"rendered":"Milestone Projects"},"content":{"rendered":"\n
\"\"<\/figure>\n\n\n\n

<\/p>\n\n\n\n

This page highlights my most significant academic projects completed throughout my Cybersecurity career at Old Dominion University (ODU). Each project reflects hands-on application of core security principles \u2014 spanning programming, network infrastructure, Linux administration, and research analysis. Use the links and resources below to explore source code, full technical reports, and supporting documentation.<\/p>\n\n\n\n

\n\n\n\n

\ud83d\udd2c COVA \u2014 Multi-Agent LLM Framework for Conversational Smishing Detection<\/h2>\n\n\n\n

Mentor: Dr. Ayan Roy, Department of Computer Science, Christopher Newport University<\/em>
Origin: Coastal Virginia Center for Cyber Innovation (COVA CCI) \u2014 ODU Node, Spring 2026 cohort<\/em><\/p>\n\n\n\n

\n

COVA (Cognitive Operations Virtual Assistant) is a multi-agent LLM framework I developed to generate synthetic multi-turn scam conversations targeting elderly populations, supporting research into conversational smishing detection that would protect vulnerable people before harm is done. The work began in Spring 2026 through the COVA CCI Cybersecurity Undergraduate Research Program at Old Dominion University, where it produced the first published paper. It has since continued independently into a follow-up paper (COVA-X) extending the dataset, generation pipeline, and detection benchmarks. The long-term direction is a fully autonomous AI agent that can identify, engage, and neutralize scam calls at point of contact and in real time \u2014 whether the threat is a human caller running a script or an AI-driven bot doing it at scale. <\/p>\n<\/blockquote>\n\n\n\n

\ud83d\udcc4 Paper 1 \u2014 A Synthetic Conversational Smishing Dataset for Social Engineering Detection<\/h3>\n\n\n\n

Produced under the COVA CCI Spring 2026 program \u2014 currently under double-blind peer review at IEEE CNS 2026 (decision anticipated late July 2026)<\/em><\/p>\n\n\n\n

This paper introduces COVA \u2014 a multi-agent LLM framework that generates synthetic multi-turn scam conversations across eight elder-targeted scam categories. The work establishes a labeled dataset of 3,201 conversations and benchmarks eight detection models spanning traditional ML and transformer architectures. XGBoost + TF-IDF achieved 72.5% accuracy as the strongest single-model baseline on the 3-class outcome prediction task.<\/p>\n\n\n\n

\ud83d\udd17 arXiv preprint<\/a>
\ud83d\udd17 ResearchGate version<\/a>
\ud83d\udd17 COVA CCI Spring 2026 cohort (Cybersecurity Undergraduate Research)<\/a><\/p>\n\n\n\n

\ud83d\udcc4 Paper 2 (COVA-X) \u2014 An Expanded Synthetic Conversation Dataset for Multi-Turn Smishing Detection<\/h3>\n\n\n\n

Independent continuation of the COVA research \u2014 submitted to IEEE BigData 2026 for peer review<\/em><\/p>\n\n\n\n

This follow-up research expands the dataset to 10,985 labeled conversations through an improved generation pipeline that addresses contamination, label mismatch, and stage-direction artifacts encountered during earlier iterations. Key findings include:<\/p>\n\n\n\n