I am a research professional specializing in optimization, data science, and machine learning. This portfolio provides an overview of my key projects, including links to publications and code.
At USMA, I researched methods to improve the accuracy of complex variable boundary element solutions to the Laplace equation. My work demonstrated that specific node positioning algorithms could increase accuracy by an order of magnitude, using groundwater flow modeling as a proof of concept.
- Poster Modeling the Flow Around a Constrained Circular Obstruction
- Published Paper Using a Node Positioning Algorithm to Improve Models of Groundwater Flow Based on Meshless Methods
My undergraduate thesis focused on evaluating admissions data letters of recommendation. Previous methods were imprecise, subjective, and unintentionally disadvantaged specific groups of applicants. Faced with this challenge, we implemented statistical NLP methods to analyze several past decades of data, eventually forming the basis of a more objective and equitable admissions process.
- Unpublished Paper The Hidden Message in Letters of Recommendation: A Natural Language Processing Analysis of Letters of Recommendation at the United States Military Academy
- Published Paper Can the Participant Speak Beyond Likert? Free-Text Responses in COVID-19 Obesity Surveys
My group at Lincoln Laboratory researched machine learning techniques to predict the impact and severity of cyberattacks on ICS networks for the FAA. The initial research developed a simulated ICS environment along with metrics for cyberattack severity based on foundation models such as MITRE ATT&CK. I spearheaded the implementation of reinforcement learning agents that optimally scan and exploit the ICS environment like a penetration tester.
NOTE: The project is classified, but the following repository shows some initial open source methodologies used to create a cyberattack environment in OpenAI Gym and scan the simulated network.
- GitHub Repository Link
My master's thesis improved traditional cyber risk ratings by incorporating features from a company's digital supply chain. I first used unsupervised learning to establish the link that companies with more extensive supply chains were at a substantially higher risk for cyberattacks. I then developed a scalable cyber risk model, framed as a Markov Decision Process, and trained an "attacking agent" using deep reinforcement learning to simulate network attacks. The resulting model was validated against data from over 13,000 companies, showing a significant improvement in cyberattack detection.
- Published Paper Innovative Supply Chain Cyber Risk Analytics: Unsupervised Clustering and Reinforcement Learning Approaches
My current work at AI2C focuses on objectively evaluating VLM-generated BDAs using open-source data, as well as developing a self-contained and portable environment to create VLM-generated BDAs on edge devices operating within research environments.
- GitHub Repository (Work In-Progress) Link