Home Recent Research People Papers Contact
Maria Ramirez
Maria Ramirez
PhD Student
Semantic Instability, Data Science
Dept. of Intelligent Systems and Robotics, University of West Florida

Background

I am a PhD student in Intelligent Systems and Robotics at the University of West Florida. I earned my master’s degree in Data Science from Johns Hopkins University and my bachelor’s degree in Mechanical Engineering from UWF. I have spent more than six years working as a civilian operations research analyst for the Department of the Air Force, at Air Force Special Operations Command (AFSOC) and the Air Force Test Center (AFTC). My professional projects have included forecasting pilot inventories, simulating flying-hour requirements, and building applications that help people plan and manage resources. I have also led collaborative work through the NATO Science and Technology Organization (NATO STO), bringing together researchers from nine NATO nations to assess defense capabilities using data and mathematical models.

I have participated in Air Force and Space Force hackathons as a developer and team lead, working on flying-hour simulations, machine learning for space-debris trajectory prediction, and other operational problems. I enjoy the hands-on process of building something, testing it, and figuring out why it works or why it doesn’t.

Research

I am interested in how large language models, and AI in general, lose track of meaning during extended interactions. A response can sound reasonable on its own while moving the conversation further from the original question or intent. I am currently working on an extensive literature review that examines how researchers describe and measure this behavior across topics such as semantic drift, context loss, and inconsistency. I am developing a scoping review and classification framework to organize that literature and identify questions that need closer study, particularly how these changes accumulate over time.

I have worked with clinical data to study how changes in patient conditions affect predictions of venous thromboembolism (VTE) risk. In this project, I developed models that update risk estimates as new clinical information becomes available, then examined which changes in care were associated with rising predicted risk. I also investigated how the information driving the models’ predictions changed relative to documented care throughout a hospital stay.

In both projects, the focus was on studying what happens over time: how a model’s responses shift during a conversation, or how a patient’s risk estimate changes as new measurements arrive. Understanding those changes is a central part of my research.