Home Recent Research People Papers Contact
Satyajit Movidi
Satyajit Movidi
PhD Student
Semantic Equivalency, RAG Systems, Personalization
Dept. of Intelligent Systems and Robotics, University of West Florida

Background and Interests

I am a graduate student in Data Science in the Department of Mathematics and Statistics and a Graduate Research Assistant in the Department of Intelligent Systems and Robotics at the University of West Florida. My research interests are mainly around intelligent systems, large language models, retrieval-augmented systems, and decision support. Over time, my research has become more focused on one main question: how do intelligent systems decide which information is actually important in a particular context?

A lot of my recent work has focused on understanding how context changes the way a system behaves. Just because a system has access to more information does not always mean that it will make a better decision. What matters is how the system interprets that information, what it considers relevant, and how its behavior changes when the context changes. Because of this, I have become more interested in understanding why a system behaves differently under different conditions rather than only asking whether one system performs better than another.

Research and Interests

I have explored these questions through projects involving personalized academic advising, retrieval-augmented generation, evolving knowledge bases, and AI governance. One of these projects is AiVisor, a personalized advising system that we developed to study how personalization can influence the behavior of an intelligent system. You can read more about it here. This project played an important role in shaping the way I think about evaluating intelligent systems.

That interest later expanded into understanding what happens when the information available to a system itself begins to change. This led to DARMA, an experimental framework developed to study how retrieval systems respond when the meaning and relationships within a knowledge base evolve over time. Through this work, I became more interested in how system configuration affects the ability to recognize and adapt to changing information. DARMA has also given me opportunities to present and discuss our research beyond the lab, including a presentation at the Conference of Florida Graduate Schools at the University of Florida, as well as invited talks for Deloitte and Jackson Health System. More recently, I have been working on TwinPath AI. TwinPath looks at healthcare AI governance from a different perspective by studying whether predictive AI behavior remains aligned with the evidence and decisions observed in real-world workflows. You can find out more about TwinPath AI here. Working on TwinPath AI has further expanded my interest in how intelligent systems should be evaluated after they are introduced into complex decision-making environments, where good performance alone may not tell the complete story.

More broadly, even though much of my current work involves large language models and retrieval-based systems, I do not want my research to be limited to one particular model or technology. I am interested in intelligent systems that can work with complex and changing information, adapt when their environment changes, and make reliable decisions in real-world settings.

My long-term goal is to contribute to research that helps us better understand how intelligent systems use context, how that context changes their behavior, and how those changes can be evaluated reliably.