Spring 2026 C-Day Program

* Project will be featured during the Flash Session

• UC-011-171 Beyond Postseason Awards: Predicting Accolades via Real-Time Control Chart Signals in the NCAA Transfer Era (Undergraduate Project) by Bowen, Daniel, Cyclewalla, Shayaan,
  Abstract: In the era of the NCAA transfer portal, collegiate basketball coaches face the critical challenge of identifying and targeting elite recruits within a condensed 15-day window. This study investigates the predictability of elite player performance by analyzing postseason award winners within the Coastal Athletic Association (CAA). Utilizing game-by-game data on player efficiency, usage percentages, and Player Efficiency Ratings (PER), we implemented an Exponentially Weighted Moving Average (EWMA) control chart—a technique from the Statistical Process Control (SPC) family—to monitor performance signals. Our results indicate that the EWMA model successfully identifies future award-winning players after an average of only 8.58 games. By providing a 22-game lead before the season’s conclusion, this model offers coaches a significant strategic advantage, allowing for the early identification of potential transfer targets before the portal officially opens. These findings suggest that SPC-based modeling is a highly effective tool for predictive sports analytics and carries broad applications for future collegiate recruitment strategies.
  Department: SDSA
  Supervisor: Prof. Michael Frankel; Dr. Austin Brown
  Poster

• * UC-082-216 AI Driven Guest Support for VacationsForYou (Undergraduate Project) by Grogan, Cassidie, Elison, Kendal, Dulcio, Benjamin, Begashaw, Ezra, Faltz, Wilfred, Faltz, Wilfred
  Abstract: The AI Guest Support Assistant is a proof-of-concept, web-based chatbot designed to streamline guest support for a high-volume vacation rental operation. The system leverages a Large Language Model (LLM) combined with a Retrieval-Augmented Generation (RAG) approach to deliver accurate, context-aware responses to common guest inquiries, such as reservation details and check-in times. Built using a React frontend and a FastAPI backend, the platform integrates securely with the StreamlineVRS property management system. The solution aims to reduce call center workload by automating repetitive inquiries while maintaining a clear escalation path for more complex requests. This project evaluates the feasibility, usability, and effectiveness of AI-assisted customer support in a controlled environment.
  Department: SWEGD
  Supervisor: Prof. Yan Huang
  Poster

• UC-086-235 Student Performance Pattern Mining (Undergraduate Project) by Arevalo Colocho, Cesar
  Abstract: This project applies data mining techniques to explore patterns in a student performance dataset. The analysis focuses on discovering natural groupings of students and frequent associations among academic, social, and lifestyle attributes. Clustering and association rule mining are used to identify meaningful structures in the data, emphasizing pattern discovery and interpretation rather than outcome prediction.
  Department: CS
  Supervisor: Prof. Christopher Regan
  Poster

• UC-087-236 Early Prediction of Player Performance (Undergraduate Project) by Freeman, Grady, Truong, Hien, Mayo, Jackson,
  Abstract: This study examines whether early-season performance metrics can support player evaluation under the NCAA’s shortened transfer window. Using data from Conference USA and the Mid-American Conference, we modeled offensive (UASE) and defensive (DAR) efficiency with multiple predictive methods. Across both full-season and 9-game datasets, DAR was more predictable, with higher R² and lower RMSE values. Linear Regression consistently performed best for DAR, while KNN and Random Forest performed best for UASE depending on the dataset. Results show that meaningful performance patterns can be identified early in the season, even with limited data. These findings suggest analytics can help programs make faster, more informed decisions for recruiting, roster planning, and player retention.
  Department: SDSA
  Supervisor: Dr. Austin Brown, Prof. Michael Frankel, Dr. Muhammad Imran
  Poster

• UC-097-186 Nudox - Compiler Based Information Retrieval (Undergraduate Project) by Slabysh, Mikita, McCrary, Robert, Wirht, Miles, Ferguson, Lindsey,
  Abstract: Nudox is a language-agnostic, version-aware documentation and search platform backed by compiler-level analysis. By lowering source code to intermediate representations, Nudox extracts structural metadata, like function signatures, types, and modules independent of the source language. At the core of the platform is a custom search engine built around versioned knowledge: queries resolve to graph nodes and expand outward along structural edges using semantic heuristics, surfacing contextually relevant symbols rather than flat text matches. The result is a canonical, automatically generated source of truth that tracks how a codebase evolves across commits.
  Department: CS
  Supervisor: Prof. Sharon Perry
  Poster

• UC-098-187 Zero-Inflated Poisson Modeling of NCAA Postseason Awards (Undergraduate Project) by Lane, Charles, Bresko, Kyle, Treang, Kaleb,
  Abstract: Our project focuses on predicting postseason awards for NCAA Men's College Basketball which can be difficult to model given that less than 15% of players in a given season win awards. After evaluating basic models, we selected a Zero-Inflated Poisson (ZIP) model to account for most players receiving zero awards. We identified free-throw attempts as being the best predictor for the structural zeros present in who can win an award. The final ZIP model produced better evaluation metrics than other basic models. Accounting for structural zeros allowed us to better model how on court statistics can translate into postseason awards.
  Department: SDSA
  Supervisor: Dr. Brown, Prof. Frankel, Dr. Imran, Dr. Karami
  Poster

• UC-099-189 Spectre (Undergraduate Project) by Tobal, Alexander, Higgins Jr, Chris, Reeves, Jaylin, Leichter, Logan,
  Abstract: Spectre consists of four levels, where players complete various objectives and fight off ghosts while doing so. Our tutorial level introduces players to the mechanics, such as shooting, rear view mirror shooting, walking and jumping. With the rest of the levels focusing on completing objectives in order to progress. The final level culminates in a boss fight, ending the journey. While players explore and complete objectives, enemies drop a currency that players can spend to obtain upgrades. Getting hit by enemies not only reduces the players’ health but also applies debuffs to them making players more cautious of their surroundings. These debuffs include lower movement speed, locking out jumping, deactivating the rearview mirror, etc.
  Department: SWEGD
  Supervisor: Dr. Joy Li
  Poster | More Information

• UC-115-161 Wayward Stray: Selix (Undergraduate Project) by Tinoco, Arly, Ercole, Tyler, Stansel, Ivy, Lothman, Austin, Charland-Martin, Jeremi, Charland-Martin, Jeremi
  Abstract: Wayward Stray:Selix is a 3rd person platformer which places importance on exploration and discovery. Players will take the role of Selix as they explore an arid desert, fighting off enemies and discovering items hidden around the map, which reveal more about the game world and its characters. Selix, a young dragon, is exiled from the only home he’s known, forced into a strange land in search of a new place to call his own. Along the way, he finds a companion, a small dove that aids and guides his way. Exploring these uncharted areas, Selix discovers there’s more to the world than his small view, and discovers secrets about himself as well.
  Department: SWEGD
  Supervisor: Dr. Joy Li
  Poster | More Information

• UC-117-213 Haunted Owl Hotel – A 3D Horror Maze Chase Game (Undergraduate Project) by Griffin, Carter, Egl, Rin, Redmon, Kcyana, Portillo, Jose, Nesbit, Alana, Nesbit, Alana
  Abstract: “Haunted Owl Hotel” is a horror Pac-Man-inspired, 3D maze chase game. You play as a cute owl named Sappy trying to escape the scary hotel, but suddenly your elevator breaks down. Navigate the spooky halls to collect the candles left behind on each floor to reactivate the elevator, but be careful, after grabbing each candle, the darkness left behind will follow you. Ghosts lurk around every corner hoping to make you their next victim. Descend through each floor without losing all 3 lives to escape the haunted owl hotel and win the game.
  Department: SWEGD
  Supervisor: Prof. Lei Zhang
  Poster

• * UC-121-133 Head in the Clouds (Undergraduate Project) by Osborne, Hunter, Day, Jane, Bell, Chase,
  Abstract: Head in the Clouds is a video game that puts the player in the shoes of a child with ADHD (Attention Deficit Hyperactivity Disorder). Rain, the protagonist, is told by their mother to take out the trash, but keeps getting distracted and daydreaming instead. The player must beat platforming challenges to get Rain back on task. The narrative and gameplay is meant to represent the difficulties of having ADHD.
  Department: SWEGD
  Supervisor: Dr. Joy Li
  Poster | More Information

• * UC-122-135 Mutatio Mentis (Undergraduate Project) by Alderman, Isaac, Mizell, Braden, Sutton, Collin,
  Abstract: Mutatio Mentis is a first person, narrative heavy, puzzle-lite RPG that follows the story of a renaissance era plague doctor and their attempt to alter the minds of three subjects; a gardener, a street urchin, and a priest. The narrative is set in 1637 Florence, Italy, in the wake of the Great Plague of Milan, and draws heavily from renaissance culture. Each of the three subjects have progressively more complex personal conflicts, which present through the gameplay aesthetics of each act, as the gameplay changes to reflect the problems of each subject. Our focus is on tackling mental and emotional health through empathy and nuance rather than diagnosis. To this end, different aspects of each subject’s psyche take physical form in the dreamscape environments, allowing the player to speak and interact with them directly, influencing the state of the subject themself, and, in turn, the environment around the player. The game features many puzzles, with solutions ranging from alchemy, exploration, and dialogue, to spatial and logical reasoning. The primary goal of play is to reach the core of each subject’s mind, allowing the opportunity to change their fundamental outlook on life, if the player so chooses. Either way, the consequences will follow the player until the end. Our aim with Mutatio Mentis is to make our players think carefully about how their actions might affect each patient they treat, and find their own answer to the question of how much they should influence each one.
  Department: SWEGD
  Supervisor: Dr. Joy Li
  Poster

• * UC-123-140 P15-T2 Boating Safety Game US Army Corps | Boating MVP (Undergraduate Project) by Hamza, Maryam, Hamza, Saleh, Vanwinkle, Will, Caffrey, Trevor, Akinsunmi, Tobi, Akinsunmi, Tobi
  Abstract: This project is an interactive 2D educational boating safety game developed in Unity to teach students essential water navigation and life jacket safety practices in an engaging and immersive format. Designed in collaboration with a real-world sponsor, the game simulates a dynamic boating environment where players navigate obstacles, identify hazards, and make safety decisions under time constraints. The experience integrates instructional modules, guided character narration, and a final quiz phase that reinforces knowledge through immediate feedback, scoring, and achievement-based rewards. Players learn critical concepts such as proper life jacket fit, safe boating procedures, hazard identification, and shallow water awareness. The project combines game development principles, user interface design, audio feedback systems, and educational assessment strategies to create a scalable learning tool suitable for kiosk deployment at visitor centers. By blending gamification with real safety standards, the application aims to increase engagement, retention, and real-world boating preparedness among youth audiences.
  Department: SWEGD
  Supervisor: Prof. Yan Huang
  Poster

• UC-128-145 KSU ESports Discord Server Bot (Undergraduate Project) by Gammill, Austin, Thomas, Foster, Truong, Lam, Olubajo, Jeffrey, Ahmed, Ismail, Ahmed, Ismail
  Abstract: The KSU Esports program has requested us to develop further upon the Discord Server Bot that they are currently using. This prior implementation was developed by Capstone students last year. Our project’s goal was to build upon their work, polish existing features, commands, UI/UX, and fix known bugs. We have worked on improving the bot’s matchmaking algorithms, tournament seeding logic, API integration, database persistence layer, stability, and statistics tracking. Furthermore, we have developed the UI/UX to be more user-friendly, added support for additional games, and overhauled the bot’s database logic.
  Department: IT
  Supervisor: Prof. Donald Privitera
  Poster | More Information

• * UC-131-162 Physical 8-Bit CPU (Undergraduate Project) by Hoerner, Samuel, Merchant, Aryan, Dislen, John, Day, Kyran,
  Abstract: This isn’t an app - it’s the machine behind it. A fully functional 8-bit CPU, built from scratch, turning raw signals into real computation. The system combines our custom assembly-to-run, FPGA-driven control unit (CU), discrete transistor Arithmetical Logical Unit (ALU), external memory, and 5 registers to execute programs through a fetch-execute cycle. The result is a tangible computing platform that bridges low-level digital logic with high-level system behavior.
  Department: CS
  Supervisor: Prof. Sharon Perry
  Poster

• UC-135-129 fishGame: Mobile Hyper-casual Fishing Game (Undergraduate Project) by Rousell, Lauren, Martinez, Lisbeth, Portillo, Jacob, Miller, Jacob,
  Abstract: fishGame is a strategy-driven hyper-casual mobile game that combines the accessibility of traditional mobile gameplay with the progression depth of a roguelike. The project was developed over a three-month period using Unity 6.3 and related production tools. Following the completion of an alpha build, testing was conducted through gameplay sessions and a detailed follow-up survey to gather feedback on player experience, clarity, and engagement. Results indicated that fishGame was well received, with players reporting low levels of confusion and strong replayability. These findings suggest a clear interest in mobile games that offer greater depth while preserving the immediacy and simplicity associated with the hyper-casual genre.
  Department: SWEGD
  Supervisor: Dr. Sungchul Jung
  Poster | More Information

• * UC-136-132 Gamma Guardian: Teaching about HLH (Undergraduate Project) by Rousell, Lauren, Leichter, Logan, Portillo, Jacob, Lock, James,
  Abstract: Gamma Guardian is an educational strategy game for children ages 6 to 12 that introduces Hemophagocytic Lymphohistiocytosis (HLH) and immune-system balance through interactive gameplay. The game places players inside the human body, where they use touch controls, antibody shields, and immune-response management to defend against interferon gamma, bacteria, and pathogens. Development used Unity 6.3 and followed a level-based design with educational feedback, AI-driven enemies, and a simple visual style to support learning. The result is a fully functional game with five levels, integrated educational content, and multi-platform support, exceeding the original goal of producing only a demo.
  Department: SWEGD
  Supervisor: Dr. Joy Li
  Poster | More Information

• * UC-137-237 The House Watches (Undergraduate Project) by Martinez, Lisbeth, Rousell, Lauren, McBride, Jaime, Quiroz, Karizma, Kleine, Aidan, Kleine, Aidan
  Abstract: The House Watches is a horror-puzzle game about a boy and a dog trying to reunite after mysterious supernatural creatures, called duendes, invade their home. The game consists of two levels, each with a different mode of gameplay; level one is a more traditional item-collection horror experience, and level 2 is based around puzzles that must be completed in a limited time frame. The House Watches includes original art, music, and models, as well as a dynamic gameplay system that increases the difficulty of each level over time.
  Department: SWEGD
  Supervisor: Dr. Joy Li
  Poster | More Information

• * UC-138-166 The Allies Connect Platform: Improving Access to Community Resources Through Technology (Undergraduate Project) by Holland, Sarah, Anand, Neha, Andoh, Aldrick, Diop, Yacine, Rogers, Alex, Rogers, Alex
  Abstract: Finding help shouldn’t be difficult, but for many people, it is. Important information about food, shelter, and local support is often scattered across different websites, social media pages, and documents, making it hard to find what’s needed, especially in urgent situations. The Allies Connect platform was created to bring that information into one place. It is a centralized, mobile-friendly platform that allows users to: • Search for resources • Register for events • Connect with nonprofits At the same time, the platform also provides organizations with simple tools to keep their information accurate and up to date. By focusing on clarity and ease of use, Allies Connect helps people find the support they need more quickly and confidently.
  Department: IT
  Supervisor: Prof. Donald Privitera
  Poster | More Information

• UC-139-175 Agentic debugger and documenter (Undergraduate Project) by Omodemi, Olaoluwa, Le, jade, Dietz, Preston,
  Abstract: Modern software development teams routinely introduce subtle bugs — mutable default arguments, bare exception handlers, insecure eval()/exec() calls, resource leaks, hardcoded secrets — that escape manual review but accumulate into technical debt and security risk. Existing linters identify problems but leave remediation to the developer. This project investigates whether a coordinated multi-agent system, combining rule-based static analysis with generative LLM reasoning, can autonomously detect, fix, and document such issues with no developer involvement beyond providing the input file.
  Department: CS
  Supervisor: Prof. Sharon Perry
  Poster | More Information

• * UC-142-177 Multiplayer SPSU Tub Racing Video Game (Undergraduate Project) by Pitts, William, Urvan, William, Powell, Thomas, Young, Joshua,
  Abstract: Multiplayer Bathtub Racing revives a well-known Southern Polytechnic State University tradition through a digital multiplayer experience built in Unity. The project extends a prior single-player tub racing game by adding online multiplayer
  Department: CS
  Supervisor: Capstone Professor: Prof. Sharon Perry; Industry Mentor: Shaun Sheppard
  Poster

• UC-143-178 Georgia Laws of Life CRM Implemenetation (Undergraduate Project) by Sternon, Nicholas, Flores, Josh, Scales, Jessica, Bhuiyan, Opurbo, Patel, Shiv, Patel, Shiv
  Abstract: This project focuses on implementing a Customer Relationship Management (CRM) system for Georgia Laws of Life using the Little Green Light (LGL) platform. The organization previously relied on spreadsheets, which caused issues such as duplicate records, inefficient reporting, and difficulty managing relationships. To address this, the team analyzed existing workflows and developed a structured data model. The system was configured, and sample data including constituents, donations, schools, and contracts was successfully imported to validate the design. The results show that the CRM system improves data organization, enhances relationship tracking, and provides a more efficient and scalable solution for managing organizational data.
  Department: IT
  Supervisor: Prof. Donald Privitera
  Poster

• UC-144-182 Paracosm (Undergraduate Project) by Friday, Va'Quez
  Abstract: Paracosm is a gothic horror game revolving around a tattoo artist, named Villain, who discovers that his art has suddenly come to life. Despite this phenomenon, Villain insists on finishing his tasks before three Am, due to his superstitious nature. What he doesn’t know yet is that if he doesn’t finish by that time, then he will be forever stuck in his shop with no escape. He will also learn that not all of his lively drawings are friendly, and that the not-so friendly drawings of his will stop at nothing to make sure he fails-Knowing that if he fails, he will be forced to continue bringing his imaginary world to life and be bound to it for eternity.
  Department: SWEGD
  Supervisor: Dr. Joy Li
  Poster

• * UC-149-185 AI-Assisted Media Organization and Intake System (Undergraduate Project) by Wu, Meilun, Branham, Nevaeh, Higgins, Isaiah, Kangni, Claude, Ahmed, Fatima, Ahmed, Fatima
  Abstract: This project includes the design and validation of a metadata-driven media intake and organization system implemented for the Office of the District Attorney, Cobb Judicial Circuit. The institution produces a considerable volume of photographs and video content via outreach programs and community interaction initiatives. Yet, the organization does not have an organized framework for managing media files, which results in inefficient file retrieval and data loss over time due to a lack of standardized organizational practices. In this regard, this project aims to develop a workflow-based system for the efficient organization of media files that will be powered by a local user interface and backed up by an intake process on the backend. First, the system will automate the metadata extraction process, classify all files based on events and the date range of their creation, and provide the necessary guidance on filling out the questionnaire to apply metadata consistently. Next, the optional AI-based tagging mechanism will help improve the metadata content by providing recommendations, ensuring that humans double-check the suggested changes. The system will allow integration with the Microsoft 365 SharePoint platform and its metadata management features to facilitate efficient media searching and filtering. Considering the access restrictions, the proof-of-concept was implemented in a mock SharePoint environment.
  Department: IT
  Supervisor: Prof. Donald Privitera
  Poster

• * UC-151-197 Nest: An AI-Powered Transition Navigator for Aging-Out Foster Youth in Georgia (Undergraduate Project) by Sookra, Stephen, Delaney, Tylin, Bryant, Brenden,
  Abstract: Each year, approximately 600–700 young people age out of the Georgia foster care system with no permanent family, no housing plan, and no clear guide beyond a 250‑page state transition PDF. The outcomes are severe: high rates of homelessness, low college completion, and unstable employment. Nest is an AI‑powered, mobile‑first web application that turns this overwhelming bureaucracy into a personalized 90‑day transition plan generated in under 60 seconds. Through a short conversational intake, the system collects a youth’s age, county, housing status, and education or work goals, then uses a deterministic rules engine to determine likely eligibility for key programs such as Extended Youth Support Services (EYSS), Chafee Education and Training Vouchers, former‑foster‑care Medicaid coverage, SNAP, and KSU’s ASCEND program. A Retrieval‑Augmented Generation (RAG) pipeline grounded in a curated Georgia‑specific resource database explains each recommendation with citations and step‑by‑step actions, reducing hallucination risk. The prototype is implemented with a React/Tailwind frontend and FastAPI backend and is explicitly designed around trauma‑informed UX principles so that transition‑age foster youth receive not just information, but an actionable, deadline‑driven plan and a “caring adult in their pocket.”
  Department: CS
  Supervisor: Prof. Richard Gesick
  Poster

• * UC-152-211 CCSE Capstone Meeting Intelligence Platform (Undergraduate Project) by Crose, Calvin, Enyart, Noah, Johnson, Marcus, Jones, Jaeden, Smith, Jonah, Smith, Jonah
  Abstract: The CCSE Capstone Meeting Intelligence Platform is a web-based monitoring system designed to assist CCSE leadership in overseeing a large variety of capstone projects. Currently, the CCSE leadership handles the overseeing of projects manually, either with advisors sitting in on student-client meetings or by reviewing recordings, which can lead to problems where potential red flags are unidentified. To help manage this, the system processes Microsoft Teams meeting transcripts and analyzes them using a Large Language Model (LLM) combined with Retrieval Augmented Generation (RAG) techniques. It then will identify potential project risks like scope creep, conduct concerns, and deviations from project specifications, providing automated analysis and early risk detection to help monitor capstone projects without requiring manual review of every meeting.
  Department: SWEGD
  Supervisor: Dr. Yan Huang
  Poster

• UC-158-205 ScrappyFin — IND, Simulated Digital Wallet & Fraud Detection Platform (Undergraduate Project) by Davis, Logan, Galvan, Dante, Jordan, Zahaira, Bracey, Avery, Pham, Chris, Pham, Chris
  Abstract: ScrappyFin is an educational FinTech platform developed in partnership with The Home Depot to simulate a digital wallet system and demonstrate fraud detection techniques. Users can create virtual wallets and perform synthetic transactions in a controlled environment, enabling analysis of financial behavior without real risk. The platform combines rule-based logic with machine learning models to identify suspicious activity, such as unusual transaction amounts or patterns. An admin dashboard provides clear explanations for flagged transactions. ScrappyFin showcases how modern financial systems detect risk while serving as a practical learning tool for software engineering and data-driven applications.
  Department: SWEGD
  Supervisor: Dr. Yan Huang
  Poster

• * UC-162-194 Smart Soil Analyzer (Undergraduate Project) by Florez Garcia, Samuel, Johnson, Edward, Gamino, Aaron, Burton, Tassha, Kinney, Wyatt, Kinney, Wyatt
  Abstract: The Smart Soil Analyzer is a machine learning-based application designed to maximize agricultural efficiency and sustainability. Our team developed a predictive system using a K-Nearest Neighbors (KNN) classifier trained on a comprehensive crop recommendation dataset. The tool allows users to input key environmental and soil metrics, including Nitrogen (N), Phosphorus (P), Potassium (K), temperature, humidity, pH levels, and rainfall. By processing these variables, the model accurately predicts the most suitable crop for the specific land conditions. This solution provides farmers with data-driven insights to optimize yields, reduce fertilizer waste, and combat soil degradation through precise crop matching.
  Department: CS
  Supervisor: Prof. Lingyan Wang
  Poster

• * UC-164-215 HootNest: AI-Powered KSU Student Assistant (Undergraduate Project) by Washington, Tabitha, Steele, Aspen,
  Abstract: HootNest helps prospective Kennesaw State students get clear, reliable answers about college life. It is designed for students who may not have easy access to counselors, mentors, or campus visits. The chatbot allows students to ask the chatbot anything they need to know.
  Department: CS
  Supervisor: Prof. Sharon Perry
  Poster | More Information

• * UC-167-222 SECU Horizon (Undergraduate Project) by Duarte, Julian, Boecker, Chance, Martin, Adam, Collins, Rylan, Haggard, Oliver, Haggard, Oliver
  Abstract: Secu Horizon is a fast-paced, stealth-based 3d action platformer, where you run around in a dystopian city. The main mechanic of the game revolves around a knife projectile that the player throws and teleports to. The game will have a strong mix of smooth platforming, soft stealth sections, and bullet time combat, as the player throws around the knife to defeat enemies. The feeling that this game will invoke is that of being an unstoppable rebel ninja, in the cold dead of the night.
  Department: SWEGD
  Supervisor: Dr. Joy Li
  Poster | More Information

• * UC-177-226 Scrapper Kinetics LLC (Undergraduate Project) by Rudenko, Mikhail, Torimoto, Andrew, Lock, Jimmy, Cheng, Marco,
  Abstract: Scrapper Kinetics LLC is a multiplayer and multimodal physics puzzle game. Where players get to choose between playing in VR or Desktop mode, and then, with up to 7 friends (8 players total), try to make a profit in the harsh dead space hulks they have been hired to scrap. We made the game as a test to see how easy it is to have completely different devices interact in the same play space.
  Department: SWEGD
  Supervisor: Dr. Joy Li
  Poster | More Information

* Project will be featured during the Flash Session

• * GC-113-125 Carter’s Lake Visitor Center Boating Safety Game (Graduate Project) by Blake, Hunter, Brown, Chancelor, Robbins, Lauren, Bryant, Will, Bryant, Kendrick, Bryant, Kendrick
  Abstract: The Boating Safety Game is an educational, kiosk-based touchscreen game created for the U.S. Army Corps of Engineers and the Carters Lake Visitor Center. It is designed to improve the knowledge and engagement of boating safety concepts for visitors, particularly for students and youth. The project was developed using multiple game scenarios meant to reinforce safe boating practices through tutorial scenes, top down navigation, life jacket and required item selection tasks, and player motivation through quizzes, feedback, scores, and a star ranking system. The game’s design emphasizes accessibility and retention through simple touchscreen interaction, guided instruction, and repeated feedback on player choices. Together, these elements demonstrate a structured and engaging approach to teaching core boating safety concepts in a format aligned with the educational and operational needs of the visitor center.
  Department: SWEGD
  Supervisor: Dr. Reza Parizi
  Poster

• * GC-119-134 Pipeline for VR Embodied Lecture Authoring and AI Gesture Refinement (Graduate Project) by Aiyatham Prabakar, Rishi Kiran
  Abstract: While Virtual Reality (VR) offers immersive educational opportunities, its pedagogical success relies heavily on a genuine sense of "instructor presence". This project presents a hybrid pipeline that automatically refines presenter 3D avatar gestures using semantic AI. Our non-VR recording system captures high-fidelity facial tracking and MediaPipe for upper-body pose estimation via standard RGB video. For emotion recognition, a local Large Language Model analyzes audio transcripts to generate a timestamped emphasis track. This semantic engine, intelligently exaggerating gestures during critical lecture moments. The captured motion and AI-enhanced gestures are synthesized and replayed on a virtual lecturer within an VR environment for evaluation.
  Department: SWEGD
  Supervisor: Dr. Sungchul Jung
  Poster

• GC-126-148 Allies Connect- Georgia's Nonprofit and Volunteer Coordination Platform (Graduate Project) by Calhoun, Molly, Banks, Takeshia, Castro, David, Hanrahan, Ryan, Davis, Tarik, Davis, Tarik
  Abstract: Georgia's nonprofit services face an issue of discoverability. While many nonprofits have the resources to help their community members succeed, they have trouble actually connecting to members of the community that need their support. Connecting with these resources is challenging for community members because their avenues of communication are spread across the internet. Some have their own websites, some have a Facebook page where they post events, some rely on word of mouth and fliers, and others rely on phone chains to keep their community members informed. This means that community members seeking support need to be able to access all these separate services. Allies Connect aims consolidate some of these communication methods by creating a website where all of Georgia's nonprofit services can host their community information, and events. Allies connect provides easy access to a visual map that displays all resources and community events in an area, and allows users to easily find and contact the resources they need. Additionally, Allies Connect features a robust volunteer management system that allows other community members to get involved and support others in their community.
  Department: SWEGD
  Supervisor: Dr. Reza Parizi
  Poster

• GC-130-160 C-Day Explorer: A Domain-Aware Platform for Discovering and Extending KSU Student Projects (Graduate Project) by Jonnalagadda, Rohan, Chapaneri, Sanketh,
  Abstract: C-Day showcases some of the strongest computing projects at KSU, but once each event ends, past work becomes scattered across semester pages, posters, PDFs, and videos, making it difficult to see long-term trends or build on prior ideas. C-Day Explorer addresses this gap with a centralized, domain-aware web platform that aggregates project records from 21 semesters of C-Day archives, KSU Digital Commons, winner pages, and YouTube presentation videos. The system organizes 1,286 projects into 11 computing domains with high abstract coverage, poster and video links, and similarity-based connections that help users quickly find related work and promising directions for extension. By turning isolated showcase artifacts into an integrated, searchable history of student innovation, C-Day Explorer helps students, judges, and faculty discover prior projects, identify patterns, and move ideas forward beyond a single showcase.
  Department: CS
  Supervisor: Dr. Bin Luo, Mentor
  Poster | More Information

• GC-140-126 Machine Learning Models for Solar Power Output Prediction: A Comparative Study with Adaptive PSO-Based Random Forest Tuning (Graduate Project) by Mahabaduge, Hasitha
  Abstract: Accurate prediction of solar power output is essential for energy scheduling, grid reliability, and efficient integration of renewable resources. Because photovoltaic generation is governed by changing atmospheric conditions, forecasting output is inherently a nonlinear learning problem. This study evaluates four machine learning models — Linear Regression, Random Forest, Multi-Layer Perceptron (MLP), and an Adaptive Particle Swarm Optimization-tuned Random Forest (Adaptive PSO-RF) — using irradiance, temperature, humidity, wind speed, cloud cover, and time-derived features drawn from a dataset of 6,738 observations. Random Forest achieved the strongest overall performance, with an RMSE of 1,816.24 and an R² of 0.9533. The Adaptive PSO-RF produced nearly equivalent results, confirming that the baseline Random Forest parameters were already near-optimal for this dataset. Feature interpretation was examined through both impurity-based and permutation importance. Whereas impurity-based importance erroneously ranked humidity as the dominant predictor, permutation importance yielded a physically meaningful ranking in which irradiance and hour emerged as the primary drivers. The study demonstrates that nonlinear ensemble methods substantially outperform linear regression for solar power forecasting and that careful interpretation of feature importance is necessary whenever predictors are correlated.
  Department: CS
  Supervisor: Dr. Coskun Cetinkaya
  Poster

• GC-141-167 Smart Resume Screening & Interview Preparation Assistant (Graduate Project) by Olien, Loreli, King, Tomas, Chen, Hanzhi, Gray, Kawanda, Frazier, Brittany, Frazier, Brittany
  Abstract: The hiring process often relies on manual resume review and keyword matching, which can lead to inconsistent and biased candidate evaluations. This project introduces a Smart Resume Screening and Interview Preparation Assistant designed to improve transparency and consistency in early-stage candidate evaluation. The system allows recruiters to upload resumes and job descriptions, then uses embedding-based semantic matching and large language models to assess candidate alignment across skills, experience, education, and projects. The application generates structured rankings, explainable insights, and tailored interview questions. This project focuses on developing a functional prototype that demonstrates how AI can enhance decision support while maintaining human control in hiring workflows.
  Department: IT
  Supervisor: Dr. Ying Xie
  Poster

• GC-154-201 Evaluation of Generative AI Responses to Pharmacy Prompts (Graduate Project) by Syed, Abrar
  Abstract: Generative AI (GenAI) is increasingly used in pharmacy for drug information and decision support, yet accuracy remains variable. We systematically reviewed studies that reported full prompts and model responses to evaluate correctness across pharmacy‑relevant tasks. GenAI performed well on basic drug facts but was inconsistent for patient-specific recommendations and interaction checking, with occasional hallucinations. Findings support cautious, supplementary use in pharmacy practice and education.
  Department: SDSA
  Supervisor: Dr. Amir Karami
  Poster

• GC-155-130 Multimodal Speech-Based Dementia Detection (Graduate Project) by Hood, Tyler, Blanco, Carlos, Shetty, Prajwal,
  Abstract: Early detection of dementia is important for timely intervention, but traditional diagnostic procedures remain costly, time-consuming, and difficult to scale. Speech-based analysis offers a promising non-invasive alternative because cognitive decline often affects fluency, articulation, and other acoustic properties of speech. In this work, we present a multimodal dementia detection framework that combines self-supervised speech representations from wav2vec2 with demographic metadata, including age, gender, and ethnicity. We first compare the multimodal approach against a strong audio-only baseline under a controlled experimental setup. We then extend the analysis with a systematic ablation study and repeated-run statistical evaluation to measure the contribution of individual metadata features. Results show that the multimodal model consistently outperforms the audio-only baseline, with the largest gain primarily driven by age. In contrast, gender and ethnicity provide only marginal independent benefit. Across repeated experiments, the multimodal configurations also show stable performance, supporting the robustness of the observed improvements. These findings suggest that self-supervised speech embeddings capture meaningful dementia-related information, while selected demographic context, especially age, can provide complementary predictive value. Overall, this work strengthens the case for multimodal learning as a practical direction for scalable speech-based dementia screening.
  Department: CS
  Supervisor: Dr. Zongxing Xie
  Poster

• GC-161-210 Hybrid Path Planning using Genetic Algorithm (Graduate Project) by Tigani, Caitlin, Joyee, Ramisa Fariha, Mohammad, Wasif,
  Abstract: This research investigates whether uninformed search (BFS) or informed search (A*) is more effective when combined with a Genetic Algorithm for maze path planning. We design and implement four algorithms: baseline BFS and A*, hybrid GA+A*, and hybrid GA(BFS+A*). Our findings show that while A* alone performs optimally, integrating it with GA can produce alternative quality solutions, though with computational trade-offs. The study demonstrates that GA+A* provides the best balance between solution quality and runtime efficiency.
  Department: CS
  Supervisor: Dr. Chen Zhao
  Poster

• GC-168-220 Active Directory to Cloud Security Data Pipeline (Graduate Project) by Butler, Michael, Shamshad, Shahiba, Touil, Mounia, Afantchao, Koko,
  Abstract: This project builds an automated pipeline that extracts identity and asset data from on-prem Active Directory, stages it in Google BigQuery, and securely sends normalized data to Lucid through Google Cloud Run. Using PowerShell scripts, the system collects users, groups, computers, DNS, DHCP, and related metadata without changing the source environment. BigQuery serves as the staging layer for validation and processing, while Cloud Run transforms and transfers the latest data to Lucid for visualization. The goal is to provide a repeatable, traceable, and reliable workflow that improves visibility into identity relationships for security investigations, auditing, and validation in a controlled sandbox environment.
  Department: SWEGD
  Supervisor: Dr. Reza Parizi
  Poster

• GC-172-139 Detection of SMS Spam using Transformer BERT Model (Graduate Project) by Bonner, Nathan, Kandell, Zachary, Hayes, Michael, Quintanilla, David, Greenberg, Leon, Greenberg, Leon
  Abstract: This project evaluates automated SMS spam classification by comparing traditional machine learning against modern transformer architectures. We built a Bidirectional LSTM (BiLSTM) baseline using TF-IDF feature extraction and NearMiss-1 undersampling to handle severe class imbalances. We then compared this against a fine-tuned Hugging Face Sentence-BERT model. Preliminary results show Sentence-BERT significantly outperformed the BiLSTM baseline (99.01% vs. 95.65% accuracy). These findings demonstrate that transformer-based embeddings offer a highly accurate, scalable solution for spam mitigation without relying on aggressive data undersampling.
  Department: CS
  Supervisor: Dr. Dan Lo
  Poster

• * GC-173-232 A Surrogate Accountability Framework for Agentic AI Systems (Graduate Project) by Tubbs, Crystal
  Abstract: Agentic AI systems introduce new accountability challenges because autonomous agents can act, adapt, and execute decisions without continuous human oversight. This research develops the Surrogate Accountability Framework (SAF), an architectural approach for embedding external oversight, traceability, and control directly within agentic workflows. To operationalize SAF, a working system, Chrysalis, was designed and implemented as a real time governance layer that monitors agent behavior, evaluates decision pressure, and enforces constraints through validation and intervention mechanisms. By shifting accountability from post hoc evaluation to continuous system level enforcement, this approach reduces the risk of compounding errors and enables interpretable, actionable control signals. The results demonstrate a practical pathway for deploying agentic AI systems with enforceable accountability in real world environments.
  Department: CS
  Supervisor: Dr. Martin Brown
  Poster | More Information

• * GC-178-191 Communication App: AI-Assisted AAC Platform (Graduate Project) by Wills, Alex, Koya, Maryam,
  Abstract: The Communication App is an accessibility-focused mobile application designed to support individuals with speech impairments, strong or unrecognized accents, and neurodivergent communication needs. The system leverages AI assisted speech-to-text (STT) and text-to-speech (TTS) technologies to enable real-time and seamless communication between users.This project aims to bridge communication gaps by providing a customizable, adaptive platform that learns user speech patterns over time. The application integrates cloud-based services, secure communication protocols, and an intuitive user interface to ensure usability, performance, and accessibility.
  Department: CS
  Supervisor: Prof. Arthur Choi
  Poster | More Information

* Project will be featured during the Flash Session

• * UR-084-219 Towards Bounding the Behavior of Neural Networks (Undergraduate Research) by Nwankwo, Emmanuel
  Abstract: Modern neural networks are typically considered black-box systems: while they are able to achieve state-of-the-art performance in many domains, it is difficult to elicit the reasons behind their decisions. From this, a sub-field of artificial intelligence called eXplainable Artificial Intelligence (XAI) arose to fill this gap. One approach to XAI is based on the symbolic compilation of a neural network's behavior to a logical formula. However, such approaches are limited in scalability, due to the fundamental difficulty of the problem. This research instead proposes an incremental and anytime approach to explaining the behavior of a neural network, for image recognition. Our approach is based on a recently published result by a KSU undergraduate, that proposed an incremental and anytime approach to explaining the behavior of an individual (threshold) neuron. To visualize the behavior of an individual neuron, we propose to enumerate prototypical examples of images that activate the neuron. We propose to visualize the behavior of a neural network by appropriately aggregating the prototypical examples of its neurons. Preliminary results suggest that such aggregate visualizations reveal interpretable patterns that can reveal the reasoning behind a neural network's decisions. Further, our approach can provide such insights in a more efficient, incremental fashion, compared to prior compilation-based methods which are by nature exhaustive.
  Department: CS
  Supervisor: Dr. Arthur Choi
  Poster | More Information

• UR-133-165 Quantum Machine Learning for Science and Engineering (Undergraduate Research) by Barnes, Barclay, Zharikov, Anna, Hamzi, Meriem,
  Abstract: Quantum machine learning (QML) has emerged as a promising method for overcoming the computational limitations of classical machine learning when analyzing large and complex data sets. This project investigates the application of QML algorithms to real-world science and engineering problems, with a focus on civil and environmental engineering datasets. We develop and evaluate a Python-based system, implemented in Google Colab, that integrates multiple quantum computing frameworks, including PennyLane, TensorFlow Quantum, and Qiskit, to implement and compare several QML models against their classical counterparts. The proposed system explores a range of algorithms such as Quantum Neural Networks, Quantum Support Vector Machines, Quantum Principal Component Analysis, and Quantum Logistic Regression across at least five engineering use cases, including structural health monitoring, traffic flow prediction, air quality, flood forecasting, and pollution modeling. Performance is evaluated using accuracy, computational efficiency, and scalability. The results highlight scenarios in which QML demonstrates a potential advantage over a classical approach, while also identifying the current limitations that come with near-term quantum devices. This work contributes to a modular framework for learning and applying quantum machine learning and provides insight into its practical viability and application within science and engineering applications.
  Department: CS
  Supervisor: Dr. Yong Shi; Prof. Sharon Perry
  Poster | More Information

• * UR-147-188 Staged Multi-Modal Alzheimer Classification using Uncertainty Quantification (Undergraduate Research) by Chen, Branden, Litton, Ethan, Doan, Long,
  Abstract: Diagnosing Alzheimer’s disease often depends on costly neuroimaging techniques such as MRIs and PET scans, which are not always accessible and can place a significant financial burden on healthcare systems. Existing clinical workflows lack a reliable way to determine which patients truly require these advanced tests, resulting in either unnecessary imaging or delayed and inaccurate diagnoses. To address this challenge, we propose the Uncertainty-Driven Dual-view (UDD) model, a multi-stage framework that integrates low-cost clinical and structural data with uncertainty-aware learning. The model first generates predictions using accessible data and quantifies its confidence, referring only high-uncertainty cases for further evaluation with expensive imaging modalities. This selective escalation strategy enables more efficient use of resources while preserving diagnostic accuracy. By combining multi-modal learning with uncertainty-guided decision making, the proposed approach offers a cost-effective and scalable solution for early Alzheimer’s disease screening.
  Department: CS
  Supervisor: Dr. Chen Zhao
  Poster

• * UR-160-200 Tortured Artist (Undergraduate Research) by Tigani, Caitlin, Scholl, Ben, Tucker, Adam, Tucker, Anaiya,
  Abstract: You are a photographer that wants to move out, so you take pictures of your house to give to your real- estate agent. However, as you are developing the photos you hear a noise that makes you turn on the lights, ruining your photos. Now you must retake the photos before morning, but something around the house has changed. Rooms are no longer in the right place, items are moved around, doors are locked, and an entity is watching you. Will you find the secrets within the puzzles or be left tortured?
  Department: SWEGD
  Supervisor: Dr. Joy Li
  Poster | More Information

• * UR-166-204 Multi-user STEM Learning Experience Powered by LLM Conversational Agents (Undergraduate Research) by Haynes, Devon, Boecker, Chance, Haggard, Oliver,
  Abstract: While Extended Reality (XR) provides experiential and interactive foundations for STEM education, current storytelling and narrative-driven applications often lack responsive nonplayer characters (NPCs), limiting interactive potential through pre-scripted stories. Additionally, despite the growth of Large Language Model (LLM) integration in XR, limited research explores the combined use of multi-user XR systems and conversational Artificial Intelligence (AI) to facilitate real-time, adaptive instruction. This project seeks to address these gaps by 1) Developing a narrative-driven STEM learning XR prototype that incorporates synchronous multi-user interaction and an embedded LLM-driven conversational agent and 2) Exploring the effectiveness of combining these technologies to improve learning engagement and educational outcomes.
  Department: SWEGD
  Supervisor: Dr. Lei Zhang
  Poster

• * UR-171-118 AidFlow: A Predictive Financial Aid Transparency System for Students (Undergraduate Research) by nakkana, chaathurya
  Abstract: Students frequently experience delays and confusion regarding financial aid refunds due to unclear system statuses and lack of communication. This project introduces AidFlow, a predictive financial aid transparency system that translates complex financial data into clear explanations, predicts refund timelines, and provides actionable guidance. A rule-based model and system pipeline were developed to simulate real-world scenarios and improve student understanding and decision-making.
  Department: CS
  Supervisor: Prof. Sharon Perry
  Poster

* Project will be featured during the Flash Session

• GRM-010-170 Machine Learning Framework for E. coli Prediction and Forecasting (Masters Research) by Devaraj, Sangeetha, Mhatre, Jui,
  Abstract: Water quality monitoring is essential for public health and environmental sustainability, yet existing monitoring infrastructures remain sparse, fragmented, and incomplete. Data from the United States Geological Survey (USGS) indicate that while over 1.5 million sites are cataloged in the USGS Water Data for the Nation, only a small fraction are actively reporting water quality measurements, with significant reductions observed in recent years. Moreover, critical parameters such as pH, water temperature, dissolved oxygen, turbidity, and microbial indicators like Escherichia coli are inconsistently measured, with widespread missing and irregular data. This work presents an AI-enabled water quality data framework designed to address these limitations by integrating heterogeneous environmental datasets, reconstructing missing observations, and enabling predictive analytics. The proposed framework incorporates spatio-temporal modeling and location-aware representations to capture upstream–downstream dependencies and environmental influences such as rainfall and watershed dynamics. Building on this foundation, we develop predictive models for E. coli concentration and contamination risk forecasting, transforming sparse and incomplete monitoring data into actionable insights. Building on this foundation, we transform sparse monitoring data into actionable insights through continuous forecasting and binary risk classification. Our proposed regression model captures the underlying environmental physics, explaining 71% of the variance (R2=0.71) in same-day E. coli loads. While continuous 24-hour advanced forecasting is limited by irregular testing data (R2=0.40), our novelty lies in deploying a binary safety classifier that successfully forecasts actionable public health hazards one day in advance with an overall accuracy of 74%.
  Department: CS
  Supervisor: Dr. Ahyoung Lee
  Poster

• * GRM-012-173 Can You Trust AI Code? Understanding and Detecting Breaking Changes using LLMs (Masters Research) by Ferdous, K M, Chowdhury, Kowshik,
  Abstract: AI-generated code is increasingly prevalent in software engineering practices, yet its reliability in preserving backward compatibility remains underexplored. This paper presents a unified study of (i) how often AI-generated code introduces breaking changes and (ii) whether large language models (LLMs) can detect such changes from commit-level diffs with explanations. We analyze 7,191 agent-generated and 1,402 human-authored pull requests from Python repositories using an AST-based approach to identify potential breaking changes. Our results show that AI agents introduce fewer breaking changes overall than humans (3.45% vs. 7.40%) in code generation tasks. However, agents show higher risk in maintenance tasks, where refactoring and chore changes introduce breaking changes at rates of 6.72% and 9.35%, respectively. To mitigate this risk and evaluate the effectiveness of the LLM-based AI agent, we developed an AI agent, PyCoReX, that can detect breaking changes from code commits. Our agent achieves a baseline F1-score of 0.82. Our findings show that commit-level LLM-based detection can support earlier and more reliable identification of breaking changes, improving the safety of agent-assisted software development.
  Department: CS
  Supervisor: Dr. Shazibul Islam Shamim
  Poster

• GRM-05-158 Spark MLlib vs Python Frameworks (Masters Research) by Klein, Austin, Tumlin, Reed, Mackey, Brandon, Asonibare, Ayooluwa, Dasari, Sri Ajay, Dasari, Sri Ajay
  Abstract: The original study evaluated Apache Spark MLlib on large datasets showing that it was able to outperform Weka in speed while also achieving similar accuracy scores. Our research builds upon this work by extending the analysis to compare Apache Spark MLlib with PyTorch. We will measure training time, speed, and accuracy to compare the results of each approach on similar hardware specifications as the original test. Original tests show promise, though we have only implemented one of the datasets. We will continue to implement the following datasets and improve metrics.
  Department: CS
  Supervisor: Dr. Dan Lo
  Poster

• GRM-06-150 Comparative Analysis of Deep Learning Architectures for Inpatient Mortality Prediction Using Time-Series Vital Signs (Masters Research) by Meurer, Jonathan, John, Andrew Philip, Udhaya, Pranay, Robinson, Nyah, Shrivastva, Dhruv, Shrivastva, Dhruv
  Abstract: This project evaluates deep learning architectures for predicting inpatient mortality using time-series vital signs derived from the MIMIC-III dataset. A baseline Long Short-Term Memory (LSTM) model was reproduced and extended with Gated Recurrent Units (GRU), Convolutional Neural Networks (CNN), and Transformer architectures. Vital signs including heart rate, respiratory rate, temperature, and systolic blood pressure were processed into fixed-length time windows for model input. Results indicate that GRU achieved the highest performance, while Transformer models underperformed due to dataset limitations. This study demonstrates the impact of model architecture on clinical time-series prediction tasks.
  Department: CS
  Supervisor: Dr. Dan Lo
  Poster | More Information

• GRM-081-207 Leveraging Non-Parametric Longitudinal Rank Sum Tests (LRST) for Robust Global Treatment Effect Estimation in Alzheimer’s Disease (Masters Research) by Shahid, Imaan
  Abstract: Parametric approaches, such as Mixed Models for Repeated Measures (MMRM), are standard in Alzheimer’s Disease (AD) clinical trials. However, these models often falter when data violates assumptions of normality or follows non-linear trajectories—common occurrences in AD due to floor/ceiling effects on cognitive scales and heterogeneous disease progression. This study evaluates Longitudinal Rank Sum Tests (LRST) as a non-parametric alternative to maintain statistical power and robustness.
  Department: SDSA
  Supervisor: Dr. Dhrubajyoti Ghosh
  Poster

• GRM-083-218 WISE: Whitebox Importance-based Subnetwork Extraction and the Privacy-Preserving Properties of Model Compression (Masters Research) by Pederson, Mason
  Abstract: WISE (Whitebox Importance-based Subnetwork Extraction) is a structured compression algorithm which extracts task-specific subnetworks by instrumenting a pretrained networks with learned gates on transformer components and optimizing on task loss and L0 sparsity regularization. WISE maintains high task performance at high sparsity levels (81-88% accuracy at 85%) where other SOTA methods collapse to near random chance. We present the first evaluation of model compression along privacy dimensions: attribute inference resistance, training data memorization, and extraction attack vulnerability. Structured compression via learned gates produces subnetworks with favorable privacy-utility balance without any explicit privacy mechanism. WISE masks also transfer to fresh models with recovery ratios exceeding 1.0, supporting the Lottery Ticket Hypothesis
  Department: CS
  Supervisor: Dr. Md. Abdullah Al Hafiz Khan; Dr. Chih-Cheng Hung
  Poster

• GRM-094-176 Influence of Speech Disfluencies and Prompt Optimization on LLM-Based Alzheimer's Detection (Masters Research) by Arshad, Muhammad Awais
  Abstract: This study evaluates how speech disfluencies and prompting strategies impact LLM-based Alzheimer’s Disease (AD) detection. We compared transcripts with preserved disfluencies (ADReSS) against clean transcripts (ADReSSo) using four state-of-the-art LLMs. Key Discovery: Complex prompts induce a "mirror-image" classification bias, where DeepSeek models severely over-classify AD, and GPT-5.2 over-classifies Cognitively Normal (CN) individuals. Optimization Fix: Applying DSPy MIPROv2 effectively mitigated bias in simpler prompts, while TextGrad successfully optimized complex, multi-step prompts.
  Department: SWEGD
  Supervisor: Dr. Seyedamin Pouriyeh
  Poster

• GRM-095-230 A Multimodal LLM Framework for Automated Construction Blueprint Analysis with Real-Time Decision Support (Masters Research) by Arshad, Muhammad Awais
  Abstract: This study addresses the high hallucination rates of Vision-Language Models (LLMs) when analyzing complex, hybrid construction blueprints. We developed a dual-input pipeline that pairs high-resolution images with a four-layer JSON "Digital Twin" (vector text, raster OCR, geometry) to mathematically ground the LLM's visual interpretation. Key Engineering Achievement: We processed a massive 137-sheet civil engineering project with zero errors. By introducing a multi-tier JSON pruning strategy, we cut token usage by up to 70% and processed the entire batch from $13-$15 to just $0.93. Decision Support Extension: We integrated real-time traffic data (TomTom) and GDOT procedural policies to transform the pipeline from a simple document reader into a strategic project management advisor.
  Department: SWEGD
  Supervisor: Dr. Minsoo Baek
  Poster

• GRM-132-159 Integrating Causal Inference with Graph Neural Networks for Alzheimer’s Disease Analysis (Masters Research) by Peddi, Pranay Kumar
  Abstract: Deep graph learning has advanced Alzheimer’s disease (AD) classification from MRI, but most models remain correlational, confounding demographic and genetic factors with disease-specific features. We present Causal-GCN, an interventional graph convolutional framework that integrates do-calculus-based back-door adjustment to identify brain regions exerting stable causal influence on AD progression. Each subject’s MRI is represented as a structural connectome where nodes denote cortical and subcortical regions and edges encode anatomical connectivity. Confounders such as age, sex, and APOE4 genotype are summarized via principal components and included in the causal adjustment set. After training, interventions on individual regions are simulated by severing their incoming edges and altering node features to estimate average causal effects on disease probability. Applied to 484 subjects from the ADNI cohort, Causal-GCN achieves performance comparable to baseline GNNs while providing interpretable causal effect rankings that highlight posterior, cingulate, and insular hubs consistent with established AD neuropathology.
  Department: CS
  Supervisor: Dr. Dhrubajyoti Ghosh
  Poster

• * GRM-134-128 NeuroVision: Mapping Brain Signals to Language and Visual Meaning (Masters Research) by Yellu, Siri
  Abstract: NeuroVision presents a unified framework for mapping electroencephalography (EEG) signals to language and visual meaning. Extracting semantic information from EEG remains a fundamental challenge due to its low signal-to-noise ratio, high dimensionality, and inter-subject variability. To address these challenges, we propose a multimodal representation learning framework that aligns EEG signals with both textual and visual embeddings through temporal modeling, spatial brain-region decomposition, and contrastive learning. The framework integrates self-supervised pretraining with supervised multimodal alignment to learn robust and transferable representations. Experimental results demonstrate BLEU-1 of 0.1106 and ROUGE-1 of 0.1493 for EEG-to-text generation, alongside a 52% improvement in retrieval performance (R@1: 1.38%) and high cross-modal consistency (0.9998). These results provide strong evidence that EEG signals encode modality-independent semantic structure, advancing general brain-to-meaning decoding and enabling scalable, next-generation brain–computer interface systems.
  Department: CS
  Supervisor: Dr. Sanghoon Lee
  Poster

• * GRM-153-198 SafeCircle: AI and Micro-Radar-Based Remote Monitoring for Patients with AD/ADRD (Masters Research) by RAHMAN, AWAN-UR-, Borty, Soarov, Ankolu, Gowtham,
  Abstract: Alzheimer's disease and related dementias (AD/ADRD) are irreversible and degenerative neurological conditions that severely impacts neurons, resulting in cognitive decline and memory loss. This study explores a mHealth system, including a SafeCircle iOS prototype, a novel solution that combines artificial intelligence with cutting-edge micro-radar technology. The platform offers a variety of features, including management of patient and caregiver profiles, real-time alerts in case of emergencies, emergency contact lists, one-touch SOS support, sharing of live locations, and recording of unusual events in video. It is a responsive and reliable care assistant that optimizes patient safety while reducing caregiver burden.
  Department: IT
  Supervisor: Dr. Nazmus Sakib; Dr. Sumit Chakravarty
  Poster | More Information

• * GRM-156-153 Finding Top-K assignments for Multi-Hypothesis Tracking (Masters Research) by Hood, Tyler, Gurung, Rakshak,
  Abstract: Multi-Hypothesis Tracking (MHT) is a framework for solving the data association problem in multi-target tracking by maintaining multiple possible assignments between observations and targets over time. Rather than committing to a single solution, MHT explores a set of competing hypotheses, allowing it to handle noise, missed detections, and ambiguous measurements. In practical systems such as radar, LiDAR, and vision-based tracking, MHT is commonly implemented using algorithms like Murty’s algorithm to generate multiple high-quality assignment solutions from the Hungarian algorithm. In this work, we instead propose an assignment-tree-based approach, where hypotheses are incrementally constructed and prioritized using a structured search strategy. This allows for more flexible exploration and pruning of the hypothesis space compared to traditional k-best enumeration methods.
  Department: CS
  Supervisor: Dr. Arthur Choi
  Poster

• GRM-157-180 Precision Engineering: Using AI to Design Nanoparticles that Target Malignant Cells (Masters Research) by Gurung, Rakshak, Tkabladze, Nino,
  Abstract: The challenge of predicting nanoparticle distribution remain a significant hurdle in nanomedicine. This research presents a computational framework for the inverse design of nanoparticles, utilizing ML models to optimize drug delivery systems for tumor targeting. By analyzing the relationship between nanoparticle compositions and biological accumulation, the model identifies optimal configurations to maximize therapeutic efficacy. The results demonstrate that AI-driven inverse design can significantly streamline the development of precision nanocarriers, reducing the need for exhaustive experimental trials.
  Department: CS
  Supervisor: Dr. Arthur Choi; Dr. Asahi Tomitaka
  Poster

• * GRM-159-214 Transportation Energy and Emission Modeling and Analysis Tool (TEEMAT) (Masters Research) by Neno Aloyem, Laeticia
  Abstract: The Transportation Energy and Emission Modeling and Analysis Tool (TEEMAT) is a web-based decision-support framework for evaluating the environmental impacts of EV adoption across U.S cities. TEEMAT integrates a feedforward neural network trained on MOVES 4.0 for tract-level vehicle emissions (CO₂, NOₓ, PM₂.₅), a macroscopic traffic and activity-based model capturing congestion-driven emission spikes, and a Meta-Prophet model trained on NREL Cambium data for grid emissions (CO₂, CH₄, N₂O). Results show that while EV adoption reduces tailpipe emissions, rising travel demand and congestion-induced low speeds can significantly offset these gains underscoring that meaningful decarbonization requires coordinated transportation and energy grid strategies.
  Department: IT
  Supervisor: Dr. Mahyar Amirgholy and Dr. Chenyu Wang
  Poster | More Information

• * GRM-169-137 Predicting the Stock Market's Next Move: How Neural Network Architecture Shapes Forecasting Accuracy (Masters Research) by Powell, Roderick
  Abstract: Three feedforward neural network (FFNN) architectures — bottleneck, parallel multi-path, and residual parallel — were trained on ten years of daily S&P 500 (SPY ETF) price and volume data to predict next-day market direction (Up/Down). All three demonstrated predictive ability above random chance. Architectural choice directly determined class prediction bias: the bottleneck concentrated errors on Up days, the parallel architecture distributed them evenly, and residual connections inverted the bias toward Down days. Model 2 (parallel) achieved the highest test accuracy (58.2%) and the most balanced class predictions among the three configurations.
  Department: CS
  Supervisor: Dr. Abdullah Khan
  Poster

• * GRM-170-221 Mind Modeling for Neuroadaptive VR Labs (Masters Research) by Nagulapally, Harish Chary
  Abstract: Students with motor impairments lack equal access to hands-on STEM laboratory experiences. This study investigates whether EEG signals and eye-gaze data contain distinguishable patterns linked to specific hand motor functions both physical and imagined. Identifying this relationship is the critical first step toward a BCI-driven VR system for accessible STEM education.
  Department: CS
  Supervisor: Dr. Sungchul Jung
  Poster

• * GRM-174-228 Mitigating Prompt-Induced Variability in LLM Outputs (Masters Research) by Tubbs, Crystal
  Abstract: Large language models in enterprise settings often produce structurally invalid outputs when users communicate informally, creating silent failure modes that pass unnoticed in downstream systems. This study investigates how prompt variation alone impacts schema compliance and output reliability. We evaluate three architectures across two tasks and four prompt styles, isolating the effect of interaction style on model behavior. Results show that baseline systems achieve 100% compliance under structured prompts but fail completely under ambiguous and casual inputs. A minimal reliability pipeline consisting of generation, self critique, and schema validation restores 100% compliance across all conditions at a predictable computational cost.
  Department: CS
  Supervisor: Dr. Martin Brown
  Poster | More Information

• GRM-175-233 From Leakage to Reliability in Dementia Detection (Masters Research) by Tubbs, Crystal
  Abstract: Automated dementia detection from speech offers a scalable approach to cognitive screening, but its reliability depends on rigorous experimental design. In this work, we reconstructed a Wav2Vec2-based dementia classification pipeline and identified critical methodological flaws, including speaker leakage, nondeterministic preprocessing, and invalid test partitions. We rebuilt the dataset using strict speaker-level separation, deterministic segmentation, and validation checks to ensure reproducibility. The corrected baseline achieved an accuracy of 44.74 percent and macro F1 score of 0.4439, reflecting a more realistic performance estimate than prior inflated results. This work establishes a scientifically valid foundation for evaluating augmentation strategies such as SpecAugment in future phases.
  Department: CS
  Supervisor: Dr. Zongxing Xie
  Poster | More Information

• GRM-176-225 MemoryEIL: An Enhanced Memory Layer Architecture for Heterogeneous Robots (Masters Research) by Shen, Yukang
  Abstract: Embodied agents still struggle to generalize across robot types, tasks, and environments because most policies remain tightly tied to robot-specific observations and action spaces. While recent VLA and planning methods improve task performance, they still lack a shared memory layer for storing and reusing experience across heterogeneous robotic systems. We propose MemoryEIL, a predicate-based memory layer that converts multimodal observations and execution traces into structured graph memories while preserving raw embeddings for fine-grained retrieval and disambiguation. MemoryEIL separates short-term task belief from long-term experience and plugs retrieved memories into either VLA policies or differentiable TAMP planners. Preliminary results show better retry behavior, fewer repeated failures, and more reliable execution than planning without memory.
  Department: SWEGD
  Supervisor: Prof. Yan Huang
  Poster | More Information

• GRM-179-195 A Retrieval-Augmented Generation (RAG) System for Bible Question Answering Using Scriptural Text and Commentary (Masters Research) by Koya, Maryam, Mishra, Pragya,
  Abstract: This project develops and evaluates a question-answering (QA) system designed to address theological and interpretive questions about the New Testament. It uses a Retrieval-Augmented Generation (RAG) framework that integrates a pretrained large language model with a structured knowledge base consisting of public-domain Berean Standard Bible (BSB) New Testament and New Testament commentaries. User queries are embedded to retrieve semantically relevant passages, which are then supplied as contextual input for answer generation. The system is evaluated based on retrieval quality, answer faithfulness, and comparison to ground truth. Performance is benchmarked against a baseline BM25 keyword retrieval system without commentary, demonstrating that commentary-augmented semantic retrieval improves interpretive accuracy and reduces hallucinated responses.
  Department: CS
  Supervisor: Dr. Dylan Gaines
  Poster | More Information

* Project will be featured during the Flash Session

• GRP-0100-193 Are TGNN-Based Intrusion Detection Results Trustworthy? A Dataset Audit and Evaluation Framework (PhD Research) by Chowdhoury, Faysal, Suer, Sait, Zhang, Yinning,
  Abstract: Temporal Graph Neural Networks (TGNNs) have reported near-perfect accuracy in Network Intrusion Detection (NID). However, this research reveals these results are often artifacts of dataset flaws rather than genuine model capability. Through a systematic audit of five benchmark datasets, we identify critical issues: node identity leakage, feature extraction artifacts, train/test contamination, and temporal sparsity. We demonstrate that models often learn to recognize specific attacker IP addresses instead of generalizing attack behavior. We propose a standardized evaluation framework featuring leakage-aware relabeling and attack-aware chronological splitting to provide a more reliable basis for future TGNN-NID research.
  Department: CS
  Poster

• GRP-01-196 Topological Drift Predicts Epidemic Instability (PhD Research) by Fanning, Charles
  Abstract: We study whether changes in contact-network topology predict transitions into high-risk epidemic periods across several classical empirical proximity network datasets. We use temporal graph learning with persistent homology-based topological signals and evaluate large-outbreak risk using SIR simulations to test whether topological drift serves as an early warning signal for epidemic instability.
  Department: SDSA
  Supervisor: Dr. Mehmet Aktas
  Poster

• GRP-02-199 Topological Constraints for Protein Folding (PhD Research) by Fanning, Charles
  Abstract: We study whether topological loss-based constraints improve multidomain whole-chain protein structure prediction beyond the ColabFold baseline by better preserving the topologies of folded proteins. We benchmark against Wasserstein metrics with our own virtual persistence and RKHS semi-metric constraints as well as higher-order virtual persistence diagrams.
  Department: SDSA
  Supervisor: Dr. Mehmet Aktas
  Poster

• GRP-03-141 Stress-Testing Parkinson’s Disease Screening: A Cross-Modal Analysis of Drawing and Speech Models (PhD Research) by Razzak, Rehma
  Abstract: Medical AI systems are increasingly deployed in clinical settings, yet most published models report only clean accuracy, dataset details, and training procedures—while omitting security‑critical evaluations such as robustness to perturbations, adversarial vulnerability, and failure modes under realistic noise. This project addresses that gap by building a cross‑modal robustness assessment for Parkinson’s disease (PD) screening models across handwriting trajectories, speech‑derived acoustic features, and an LLM‑based preprocessing layer. Despite strong clean performance (visual subject‑level ROC AUC ≈ 0.99; audio ≈ 1.0), the visual pipeline proved highly brittle to realistic acquisition distortions. Downsampling and point‑dropout caused near‑chance collapse, while pressure noise and XY jitter produced monotonic degradation and threshold instability. Simple defenses improved robustness in corruption‑specific ways—augmentation was the strongest general‑purpose method, and resampling+augmentation best mitigated sampling‑density failures—yet all defenses reduced clean accuracy. Small adversarial perturbations (ε ≈ 0.2–0.3) reliably flipped predictions. Speech‑feature models were naturally robust to random corruptions and generalized well across utterance types, but adversarial attacks again caused sharp collapse at small ε, revealing a shared vulnerability across modalities. An LLM layer introduced additional instability: Llama‑3 was format‑stable but less feature‑grounded, while Gemma‑3 was more accurate but more brittle. Overall, the project demonstrates that high clean accuracy does not imply real‑world reliability, and that security evaluation must become a standard component of medical AI development.
  Department: CS
  Supervisor: Dr. Michail Alexiou
  Poster

• GRP-07-163 SMISHGUARD: An AI-Powered Framework for SMS Phishing Detection and Alert System for Vulnerable Users (PhD Research) by Das, Jiban Krisna
  Abstract: This research introduces SMISH-GUARD, a multi-layer framework for adaptive SMS phishing (smishing) detection that integrates language-aware semantic modeling, graph-theoretic campaign reasoning, and cost-sensitive decision calibration within a unified architecture. The framework integrates dual transformer encoders for multilingual semantic understanding with a heterogeneous temporal graph layer that captures relational attack signals such as shared URLs, sender reuse, and campaign propagation patterns. A cost-sensitive decision optimization module is further incorporated to translate probabilistic model outputs into risk-aware alert policies that explicitly balance false-positive inconvenience against the higher societal and financial cost of missed smishing attacks. The study evaluates four integrated datasets comprising 51,528 SMS messages and 16 phishing attack subtypes under realistic class imbalance and multilingual conditions. The framework achieves PR-AUC of 0.943 and ROC-AUC of 0.950, while maintaining high recall for critical phishing subtypes. The ablation study confirms that each architectural component—including semantic modeling, graph-based risk propagation and decision calibration that contributes independently to overall robustness. Multilingual experiments also indicate promising generalization across linguistic contexts. The framework emphasizes deployment realism, incorporating probabilistic calibration, adversarial robustness considerations, and explainable decision thresholds suitable for adaptive on-device warning systems. The findings suggest that effective smishing defense requires not only accurate language models but also mathematically grounded decision policies and structurally aware representations of attacker behavior. SMISHGUARD establishes a principled foundation for next-generation smishing defense systems that combine semantic intelligence, relational reasoning, and human-centered risk optimization to mitigate evolving SMS threat.
  Department: CS
  Supervisor: Dr. Yong Shi
  Poster

• * GRP-08-168 Diagnosing Faults in Electrical Power Systems of Satellites (PhD Research) by Lasley, Jared, Thi Binh Nguyen, Nguyen,
  Abstract: Satellite systems cost hundreds of millions of dollars or more to launch. To be resistant to catastrophic failures (and total loss of investment), satellite systems are designed with redundant sub-systems and are further equipped with numerous sensors and other health-monitoring sub-systems. In this poster, we consider an approach to fault diagnosis based on probabilistic logic programming. In particular, we propose to use ProbLog to model and reason with the electrical power system (EPS) of a satellite. Once we model a system using (probabilistic) first-order logic, we can take the system state and any (unexpected) sensor readings, and through automated reasoning, we can provide ranked predictions about the most likely faults in the system. We present our approach to fault diagnosis using ProbLog, and present a case study in a simplified EPS, highlighting our ability to isolate and diagnose different categories of faults.
  Department: CS
  Supervisor: Dr. Arthur Choi
  Poster

• * GRP-09-169 Using Logic to Explain and Formally Verify the Behavior of ReLU Neural Networks (PhD Research) by Nguyen, Nguyen Thi Binh
  Abstract: Homeschooling in the United States has expanded rapidly, reaching about 3.4 million students (6.26% of K-12 school-age population) in 2024-2025, with accelerated growth following COVID-19. Understanding the reasons behind these decisions is important for informing education policy, resource allocation, and the design of schooling systems that better meet families’ needs. Most of the research on homeschooling relies on statistical methods such as logistic regression or probit models to identify significant factors associated with homeschooling decisions. In this work, we approach this research question from a different perspective by leveraging Explainable AI techniques to provide deeper insights into the homeschooling decisions.
  Department: CS
  Supervisor: Dr. Arthur Choi
  Poster

• GRP-093-174 Transforming Everyday Smartwatch Data into Clinical Early Warnings (PhD Research) by Jahan, Nursat
  Abstract: Cardiovascular Disease (CVD) related most machine learning (ML) models trained on clinical data offer high accuracy but are not practical for continuous monitoring. Smartwatch-based wearables provide continuous real-time physiological data but lack clinical validation for robust risk prediction outside the clinical setting. To bridge this gap, we proposed a novel teacher-student knowledge distillation framework that transfers knowledge of complex and large EHR datasets to a small Fitbit smartwatch dataset-based prediction model. The student model achieves promising accuracy, identifying all types of derived CVD risk profile groups. Our study introduces a non-invasive continuous health monitoring framework, demonstrating that passively collected daily metrics can be transformed to clinically powerful early warnings of an individual’s long-term cardiovascular health plan.
  Department: IT
  Supervisor: Dr. Nazmus Sakib
  Poster

• GRP-096-183 Early Warning Signals for Geopolitical Oil Shocks via Multi-Model NLP Sentiment Analysis (PhD Research) by Ohalete, Nzubechukwu
  Abstract: The Strait of Hormuz carries roughly 20% of the world’s daily oil supply. Its closure on March 4, 2026, sent Brent crude surging 36.6%, from $74.64 to a peak of $118.35/barrel. Traditional time-series models fail during such unprecedented shocks because the historical price data contains no analog. This study evaluates whether NLP sentiment analysis can detect crisis signals in news text before they appear in prices, and whether agreement patterns across models predict volatility. We score 2,249 Guardian news articles using five deterministic sentiment models across three tiers: a lexicon baseline (VADER), a general-purpose transformer (RoBERTa-CardiffNLP), and three financial-domain transformers (FinBERT, FinBERT-Tone, DistilRoBERTa-Financial). Financial-domain models detected the crisis 47 days before closure, outperforming general-purpose models by 26 days. Article volume proved the strongest volatility predictor (r=0.76, p<0.0001), while model consensus, not disagreement, signals crisis severity, inverting our original hypothesis. LSTM and XGBoost forecasting experiments show all model variants converging near a naive persistence baseline during the crisis, confirming that point prediction of unprecedented events remains fundamentally limited and reinforcing the value of upstream textual early warning.
  Department: SDSA
  Supervisor: Dr. Kevin Gittner
  Poster

• * GRP-114-127 Gamma-Sieve: Structural De-obfuscation of Financial Regime Manipulation via Heterophilic Graph Neural Networks (PhD Research) by Regan, Christopher
  Abstract: Market manipulation increasingly exploits fragmentation — dispersing orders across dozens of accounts, venues, and sub-second timing windows — to evade rule-based surveillance. We present Gamma-Sieve, a heterophilic graph neural network approach that constructs heterogeneous transaction graphs (four node types, ten edge types) from market microstructure data, applying CARE-GNN with RL-gated edge filtering and TFE-GNN with spectral triple-frequency decomposition. At production scale, heterophilic GNNs outperform a bidirectional LSTM baseline by +16% AUC on fragmented coordination attacks. However, evaluation on real NASDAQ equity data (LOBSTER Level 3) reveals a critical domain-shift challenge: GNN false positive rates of 42–88% on legitimate trading, caused by structural overlap between adversarial fragmentation and real market-making coordination. Fragmentation calibration resolves this for CARE-GNN, reducing FPR from 7.5% to 2.4%. A cross-attack experiment further qualifies the structural advantage: GNNs catastrophically fail on unseen single-agent attacks while the LSTM detects them perfectly, demonstrating that robust deployment requires multi-architecture ensembles. These results establish graph topology as essential for detecting dispersed coordination while revealing the practical necessity of domain calibration and attack-diverse training for real-world deployment.
  Department: CS
  Supervisor: Dr. Michail Alexiou; Dr. Ying Xie
  Poster | More Information

• GRP-120-138 WALL-E: Wide-Area Aerial Live Learning for Emergency Disaster Evaluation (PhD Research) by Shrestha, Shiva
  Abstract: WALL-E is an AI-powered, custom-built quadcopter that surveys disaster zones in real time, classifying building damage and generating a GPS-tagged damage map with no internet required. The system uses an RGB camera for live AI inference and a FLIR thermal camera for additional situational awareness. A custom YOLO based model runs entirely onboard the Jetson Orin Nano, logging every detection via GPS for immediate command use. Future work will expand detection capabilities to include human presence identification.
  Department: IT
  Supervisor: Dr. Honghui Xu
  Poster | More Information

• * GRP-125-144 Mapping the Affordances of Human-AI Interaction: A Large-Scale Text Mining and Statistical Analysis of LLM Usage Patterns (PhD Research) by Vallepu, Anil
  Abstract: People increasingly communicate with AI for schoolwork, office tasks, and daily needs. This study investigates the affordances of human-AI interaction using modern text mining and statistical analysis on the WildChat-1M dataset of over 1.1 million real-world ChatGPT user conversation logs. We apply BERTopic to extract latent interaction topics, compute a probabilistic topic-document matrix P(T|D), and perform rigorous statistical testing including Welch’s T-test and ANOVA to compare affordance patterns between GPT-3.5 and GPT-4.0 Results reveal that creative writing, Coding, message drafting and many interesting topics are the dominant affordances, while a spatio-temporal trend analysis maps how interaction patterns evolve globally over time.
  Department: SDSA
  Supervisor: Dr. Amir Karami
  Poster

• GRP-146-136 Platonic Policy Representations: Navigating Learned Manifolds for Rapid Adaptation (PhD Research) by Redovian, Cameron
  Abstract: Adapting reinforcement learning policies to changing dynamics is typically addressed by domain randomization, which trains a single robust policy at the cost of specialization, or by meta-RL methods, which enable rapid adaptation but require online inference or optimization. We propose a different mechanism: extending the Platonic Representation Hypothesis (Huh et al., 2024) and vec2vec (Jha et al., 2025) to policy space, we show that diverse task-competent policies trained under varying dynamics admit a shared, low-dimensional manifold structure that is learnable from trajectory embeddings. Platonic Policy Representations (PPR) learns this manifold via geometric preservation losses, then navigates it for rapid adaptation: a hypernetwork generates complete policy weights from any manifold position in a single forward pass, while a dynamics-to-manifold predictor guides deliberate exploration of qualitatively different behavioral strategies for novel dynamics configurations. On continuous control tasks with varying gravity, mass, and friction, PPR achieves 12-235% improvements over domain randomization baselines, with gains scaling with action space dimensionality: Ant (+235%, 8-dim), Walker2d (+153%, 6-dim), Hopper (+59%, 3-dim), and LunarLander (+12%, 2-dim). These results demonstrate that policy adaptation can be recast as geometric navigation of a learned manifold, offering a complementary paradigm to existing adaptation approaches.
  Department: IT
  Supervisor: Dr. Taeyeong Choi
  Poster

• GRP-148-223 Uncertainty-Guided Conservative Propagation for Robust Coronary Artery Segmentation (PhD Research) by Huang, Huan, Zhao, Chen,
  Abstract: Coronary artery segmentation plays a key role in cardiovascular disease analysis, yet existing methods often produce fragmented and structurally inconsistent vessels in challenging regions. We propose an uncertainty-guided conservative propagation (UGCP) framework that improves segmentation reliability by allowing high-confidence regions to guide uncertain ones through controlled information propagation under a conservation principle. This mechanism enhances structural continuity while preventing unstable updates. Experiments on Coronary CT Angiography (CCTA) and Invasive Coronary Angiography (ICA) datasets demonstrate improved segmentation accuracy and topology preservation. Additional evaluations on other vascular datasets further suggest the generalizability of the proposed approach.
  Department: CS
  Supervisor: Dr. Chen Zhao
  Poster

• GRP-150-192 From Construction Floor-Plan to Robot-Ready Navigation Map: A Dual-Memory Multi-Agentic AI That Learns from Its Own Successes and Failures (PhD Research) by Akhi, Amatul
  Abstract: Construction floor plans contain rich architectural information, but they are not directly usable for robotic navigation in IoT-enabled smart buildings. They often require manual processing to remove irrelevant annotations and extract navigable layouts. Existing methods either depend on fixed image-processing pipelines that do not generalize well across different floor plan styles or on data-intensive learning models that require large annotated datasets. We propose a dual-memory multi-agent framework that treats floor plan-to-map conversion as a sequential, experience-driven decision process under data scarcity. The framework uses three cooperative agents for perception, decision-making, and evaluation, which interact through a shared persistent memory represented as a relational experience graph. Each memory entry stores a visual embedding that captures floor plan characteristics and a logical embedding that records processing decisions, outcomes, and failure patterns. This design enables similarity-based retrieval, reuse of successful configurations, and explicit avoidance of past failures. Experiments on benchmark construction floor plans, open-source residential floor plans, and real-world robot navigation show strong structural preservation, stable cross-domain generalization with frozen memory, and reliable physical deployment. Our system preserves navigation-critical topology while avoiding structural corruption, demonstrating data-efficient and transferable performance across diverse floor plan styles.
  Department: IT
  Supervisor: Dr. Jian Zhang; Acknowledgment: Supported in part by the NSF Grants CCSS-2245607 and CCSS-2245608, and by a Construction Innovation and Collaboration (CCIC) grant.
  Poster

• GRP-163-229 Emotion Elicitation via Empathic AI (E-AI) Agent Teacher in VR Classroom (PhD Research) by Paul, Arpita
  Abstract: How emotions influence learning experiences in virtual reality (VR) remains unclear. To address this gap, we investigated the impact of emotionally elicited lectures on learners’ experiences and outcomes in a pedagogical VR classroom using a between-subjects design with two emotional conditions (positive vs. negative). Emotional cues were delivered by a virtual agent (VA) teacher through bodily gestures and verbal expressions during lecture delivery. In a user study (N=34), we collected multimodal data, including neurophysiological measures such as electroencephalography (EEG), galvanic skin response (GSR), heart rate (HR), heart rate variability (HRV), skin temperature, and eye gaze, along with self-reported emotion, learning experience, and quiz performance. The results showed significant differences in neurophysiological responses, particularly in EEG, GSR, HR, and temperature, across emotional conditions. However, no significant differences were observed in subjective emotional responses, learning experiences, or quiz performance. These findings suggest that emotionally expressive virtual agents can elicit measurable neurophysiological responses during VR learning, even when participants are not consciously aware of these changes, highlighting the importance of considering implicit emotional responses when designing affect-aware VR learning environments.
  Department: SWEGD
  Supervisor: Dr. Sungchul Jung
  Poster

• GRP-165-217 Evaluation of Multi-Platform Simulation Environments for Diverse Robotic Manipulation Tasks (PhD Research) by liu, Zhiguo
  Abstract: Robotic development often requires transitioning between different simulation environments to meet specific task requirements. This project presents a comparative evaluation of four major simulation platforms—Gazebo, MuJoCo, CoppeliaSim, and Isaac Sim—through the successful reproduction of diverse manipulation tasks. By implementing system integration, dual-arm coordination, sequential logic, and reinforcement learning across these engines, this study identifies the functional strengths and practical engineering constraints of each environment. The results provide a qualitative guide for selecting simulation tools based on task-specific needs, such as middleware compatibility versus physical fidelity.
  Department: SWEGD
  Supervisor: Dr. Yan Huang
  Poster

• GRP-180-147 Quantum Augmented Microgrids (QuAM) Simulator (PhD Research) by Jha, Nitin, Paudel, Prateek,
  Abstract: Small modular nuclear reactors (SMRs) are redefining the energy generation landscape by enabling the deployment of modular, scalable, and pre-built power units that can be used to build distributed autonomous microgrids for critical infrastructure and burgeoning AI factories. Often, these microgrids are linked together to provide a resilient, decentralized power generation infrastructure. Consequently, the cybersecurity of microgrids is of critical importance. In this work, we propose a quantum augmented network framework for resilient microgrids. We integrate the ideas of secure quantum networking, quantum anonymous notification, and quantum random number generation to strengthen the integrity, confidentiality, and privacy of microgrid networks. To substantiate the possible benefits of using quantum augmented microgrids, we simulate a practical high-impact classical attack: a traffic analysis and priority-action spoofing campaign that can (1) deanonymize the anonymous notification for a high-priority action, (2) force excessive key usage, and (3) induce harmful allow/block operations at the control level. We quantify how these attacks affect information leakage, spoof acceptance, key sufficiency, and operational outcomes such as latency, deadline misses, unserved energy, etc. This quantum augmented microgrid (QuAM) framework lets us evaluate trade-offs between privacy, availability, and the operational cost of mitigations (cover traffic, verification delays, and key-rotation policies), further paving the path for the study of more nuanced attacks that arise due to the use of quantum-classical integrated frameworks.
  Department: CS
  Supervisor: Dr. Abhishek Parakh
  Poster

* Project will be featured during the Flash Session

• EX-04-142 Modeling Distress and Evaluating Chatbot Safety for Suicide-Related Social Media Texts (Exploratory Project) by Razzak, Rehma
  Abstract: This project addresses the urgent need for transparent chatbot safety evaluations amid rising concerns about AI-facilitated self-harm. Using public social media datasets, we simulate two tasks: (1) detecting suicidal ideation via emotion-based risk scoring, and (2) stress-testing a support-style chatbot against 888 high-risk prompts, including euphemisms and “for a story” framing. A multi-label classifier trained on GoEmotions feeds emotion profiles into a logistic regression model to generate suicidality risk scores. These scores guide a local chatbot built with Ollama’s llama3, which analyzes user messages and steers responses toward safe, empathetic behavior. Evaluation shows ~90% of replies were safe or supportive. This framework links emotional signals to risk heuristics and enables reproducible safety testing under realistic self-harm scenarios.
  Department: SDSA
  Poster

• EX-116-209 The Understudy – A 2.5D Turn-Based Story Game (Exploratory Project) by Randolph, Ara, Egl, Rin, Herrington, Cayden, Swerdlow, Jonah, Griffin, Carter, Cruz, Amaya
  Abstract: “The Understudy” is a whimsy-filled 2.5D turn-based theatrical adventure where you play as the last-minute understudy, who has been suddenly thrust into the spotlight after the lead mysteriously vanishes right before showtime. Armed with nothing but masks (comedic, dramatic, and tragic) and a script you definitely didn’t not rehearse enough, you fight your way through a cast of dramatic acting troupe members, ranging from a painfully shy tree to a snarky jester ex to a pompous king who’s very sure you don’t belong on his stage. Swap masks to change your combat style, solve dialogue puzzles, and prove that even an understudy can steal the show when the curtain rises. Explore the set, confront your doubts, and discover the role you were never meant to play… but might be ready for anyway.
  Department: SWEGD
  Supervisor: Prof. Nick Murphy
  Poster | More Information