Events
Fall 2025 Computing Showcase (C-Day) Winners
Undergraduate Project Winners
First Place
Abstract: Our team’s project aims to enhance the Loving Arms Cancer Outreach website, making it more accessible and user-friendly for all visitors. Loving Arms is a nonprofit supporting individuals and families affected by cancer, and their website plays a key role in sharing information, connecting people to support programs, and reaching those in need. To achieve this, we first audited the current site to identify areas of confusion or outdated content, and we are currently in the process of providing a redesigned website. These updates are intended to improve the visitor experience while giving staff a reliable, easy-to-use platform that better supports the organization’s mission.
Department: Software Engineering and Game Development
Supervisor: Dr. Yan Huang
Project Sponsor: Catherine Gankofskie - Loving Arms
|
Second Place
Abstract: A 3D platformer where you can transform from a cube to a boat and a plane. The game is on mobile and features the player traversing through a vapor wave inspired level with techno music in the background. They must perform jumps and lane switches to the beat of the song, and survive to the end of the level to win.
Department: Software Engineering and Game Development
Supervisor: Dr. Sungchul Jung
Third Place
Abstract: The AI Driven Resident Inquiry Processing System is designed to enhance the National Housing Compliance (NHC) ability to process resident inquiries using artificial intelligence(AI). NHC is a 501(c)(4) not-for-profit corporation who provides training and compliance services to the affordable housing industry. Each month NHC receives over 200 inquiries from residents via phone and email. These inquiries range from general questions to urgent, life-threatening concerns. Efficiently processing and responding to these inquiries is often critical to resident safety and well being. This project uses AI to automate resident inquiries as they are received, extract and classify key information, and display this information on an interactive dashboard for close monitoring and timely resolution. By streamlining inquiry processing, this system enables NHC to respond to resident needs faster and accurately.
Department: Information Technology
Supervisor: Prof. Donald Privitera
Project Sponsors: Jeff Wirrick, Kelli Sterling
NHC IT Staff: Walter Hoang, Jamaul Morrison, Xeryus Starr, Satara Tyler
|
Graduate Project Winners
First Place
Abstract: Mass casualty triage requires quick, accurate decisions under pressure. Live training is costly and time-intensive. Virtual Reality (VR) trainings approach has shown comparable learning effectiveness compared to live trainings, motivating the use of VR simulations. This study explores how collaboration with an AI robot partner can enhance the triage training effectiveness. The Findings will contribute in understanding how human-AI collaboration enhances trainings.
Department: Information Technology
Supervisor: Dr. Hansol Rheem
|
Second Place
Abstract: A web-based platform to streamline peer evaluations in team-based courses. Professors can securely create and manage student rosters, assign students to courses/teams, trigger email invitations, and receive structured, professor- friendly reports with both numeric and textual feedback. Optional AI features may summarize comments and flag potential concerns, depending on timeline and scope.
Department: Information Technology
Supervisor: Dr. Jack Zheng
Project Advisor: Geetika Vyas
| |
Third Place
Abstract: Meetless functions as an OS for modern organizations: a secure, multi-agent runtime that turns scattered inputs (docs, emails, tickets, chats) into asynchronous, outcome-driven discussions with decisions, owners, and due dates—so teams ship without meetings. System architecture (OS metaphor).- Kernel (Orchestrator): schedules “processes” across specialized agents using routing policies, guardrails, and retry semantics. - Process I/O: unified connectors for Google/Microsoft suites, Slack, Jira/Linear, and web sources, normalized into a document graph with vector embeddings. - Memory & FS: temporal knowledge graph (Neo4j) + document store (PostgreSQL) + vector index (Weaviate) with time-aware retrieval. - Syscalls/APIs: /discussions.create, /decisions.propose, /actions.sync, /summaries.latest; all idempotent with audit trails. - Runtime & Toolchain: FastAPI services; LangGraph/LangChain for agent flows; MLFlow for experiment tracking; background workers for ingestion, scoring, and SLA enforcement. Workflow. 1. Ingest context: generate a structured Brief (goals, constraints, risks, dependencies).2. Facilitate threaded Q&A; auto-summarize positions and trade-offs. 3. Produce a Decision & Action Pack (decisions, owners, due dates) and sync to PM tools. 4. Update the temporal graph; surface follow-ups and unresolved questions. Security & governance: RBAC with tenant isolation, secret-scoped connectors, redaction/masking policies, and full decision lineage via timeline views. Evaluation targets (pilot teams, n=6): ≥40% reduction in meeting hours for standups/planning/retros; <5 min median catch-up time for late joiners; measurable increase in decision traceability and follow-up closure rate.
Department: Computer Science
Supervisor: Dennis Loubiere
|
Undergraduate Research Winners
First Place
UR-1195 Using Machine Learning to Diagnose Alzheimer's (Undergraduate Research) by ,
Abstract: Early detection of Alzheimer’s disease (AD) remains a significant clinical challenge, as the changes associated with cognitive decline are often subtle and difficult to identify through visual assessment alone. This study investigates modern machine learning methodologies to improve the prediction of cognitive impairment using volumetric MRI–derived region-of-interest (ROI) features. We constructed three binary classifiers [NC vs. AD, MCI vs. AD, and NC vs. MCI] and evaluated various algorithms, including logistic regression, random forests, neural networks, and support vector machines (SVMs). Using measurements generated from eight anatomical brain templates, our models learned patterns indicative of normal cognition, mild cognitive impairment, and Alzheimer’s disease. Among all tested approaches, the radial basis function (RBF) SVM consistently achieved the highest performance, reaching accuracies of approximately 70–80% depending on the classification task. We discuss the implications of this model’s dominance for future clinical applications and the continued development of machine learning–driven diagnostic tools.
Department: Computer Science
Supervisor: Prof. Sharon Perry
Project Sponsor: Dr. Chen Zhao
| |
Second Place
Abstract: OwlExchange is a secure, role-based campus marketplace designed to help Â鶹´«Ă˝ State University students buy, sell, exchange, and donate items in a trusted digital environment. The platform replaces unorganized and unsafe exchanges occurring across group chats and social media by providing a centralized system with authenticated user accounts, item listings, and transparent communication. Built with a Flask (Python) MVC architecture, Auth0 for secure identity management, and MySQL for persistent storage, OwlExchange supports modular dashboards for buyers, sellers, and administrators, enabling item management, interest requests, and platform oversight. By promoting reuse and donation of textbooks, furniture, electronics, and other student items, the system contributes to affordability and sustainability on campus. The project demonstrates full-stack development, secure authentication design, UI/UX branding aligned with KSU identity, and cloud deployment with a documented roadmap for future enhancements.
Department: Information Technology
Supervisor: Prof. Donald Privitera
Sponsor: Assurant Global Technology
|
Third Place
Abstract: Coronary artery disease (CAD) is one of the leading causes of death worldwide, making the assessment of blood flow and pressure distribution within coronary vessels essential for diagnosis and treatment planning. Fractional Flow Reserve (FFR) is a key measure used to determine the severity of arterial blockages, but traditional methods, such as invasive measurements or computational fluid dynamics (CFD) simulations, are not only time-consuming and costly but also invasive. This project explores the use of deep learning to predict blood pressure distribution in coronary arteries using a 3D convolutional neural network. The dataset consists of Coronary Computed Tomography Angiography (CCTA) scans paired with blood pressure obtained from CFD simulations. After preprocessing and voxelizing the CCTA scans, the VNet model learns to map the vessel’s geometry to its internal pressure distribution. Our model achieved a Pearson correlation of 0.93 and an R² score of 0.84 between the predicted and simulated pressures. These results indicate that VNet delivers accurate, spatially consistent pressure predictions that closely match CFD outputs while substantially reducing computation time. This approach underscores the potential of deep learning to accelerate non-invasive FFR estimation and enhance patient-specific cardiovascular analysis.
Department: Computer Science
Supervisor: Dr. Chen Zhao
Master's Research Winners
First Place
Abstract: The spatial organization of tumor cells, stroma, and tumor-infiltrating lymphocytes (TILs) within the tumor microenvironment plays a critical role in cancer progression and is strongly associated with clinical outcomes. However, quantifying the significance and statistical impact of these spatial patterns remains challenging due to the complex interactions among these components. In this study, we analyze spatial patterns associated with patient survival in gastric and colorectal cancer by integrating four predictive classifiers with spatial image statistics across four large patient cohorts. U-Net was used for semantic segmentation of tumor, stroma, and TILs on digitized Hematoxylin and Eosin–stained FFPE whole-slide images, while ResNet-18 was trained to predict Microsatellite Instability (MSI) status. To identify statistically significant tumor hotspot regions, we applied the Getis-Ord Gi* statistic, which evaluates local spatial relationships relative to surrounding tissue regions. Kaplan–Meier survival analyses and log-rank tests were conducted to assess associations between the spatial arrangement of tumor and stroma and overall patient survival. Our findings reveal that the stromal composition surrounding tumor hotspot regions, as delineated by the Getis-Ord Gi* statistic, is significantly associated with differences in overall survival in both gastric and colorectal cancer. Additionally, log-rank tests were used to evaluate the relationship between stromal composition, MSI status, and ACTA2 expression levels.
Department: Computer Science
Supervisor: Dr. Sanghoon Lee
Second Place
Abstract: Vision-Language-Action (VLA) systems are beginning to support everyday clinical workflows. Deltoid intramuscular injection is a representative task, but progress is limited by data scarcity, privacy constraints, and the cost of expert annotation. Recent text-to-image (T2I) models make large-scale data synthesis possible, yet ensuring anatomical correctness, diversity, and label quality remains difficult. To address this gap, we propose a Synthetic Data Engine tailored for medical perception, integrating cold-start filtering, controlled T2I generation, CLIP-based quality checks, and iterative segmentation training. We further introduce an anthropometry-grounded formulation of injection safety that produces interpretable safe-zone guidance. Experiments show that synthetic data can effectively bootstrap deltoid-segmentation performance and support reliable injection-area perception.
Department: Software Engineering and Game Development
Supervisor: Dr. Yan Huang
| |
Third Place
Abstract: NEEMAT is a web-based decision-support tool that predicts vehicle and power-plant emissions plus fuel/energy consumption under rising EV adoption for Atlanta, Los Angeles, New York, and Seattle. A feedforward neural network trained on MOVES estimates tract-level vehicle energy use and CO2/NOx/PM2.5 by speed, vehicle type, fuel, and age, while a macroscopic traffic model captures flow effects. Grid-side CO2/CH4/N2O from EV charging are forecast with a Meta-Prophet model trained on Cambium. Users can explore 24-hour profiles and five-year outlooks, compare scenarios, and export results. Findings show that despite substantial EV uptake, mixed fleets and grid responses can raise total emissions, underscoring the need for integrated transportation–power planning.
Department: Information Technology
Supervisor: Dr. Chenyu Wang
Advisor: Dr. Mahyar Amirgholy
|
PhD Research Winners
First Place
Abstract: Water quality monitoring is crucial for environmental protection, public health, and ecosystem sustainability. With increasing pressures from urbanization, agricultural runoff, and climate change, robust data-driven approaches are essential for early detection of water quality degradation and informed decision-making in environmental conservation efforts. Current water quality monitoring relies on reactive threshold exceedances, failing to detect gradual degradation and multi-parameter deterioration patterns. This creates delayed response to pollution events and missed opportunities for preventive intervention in one of Queensland's most vital water systems. The importance objective is to implement and evaluate a Real-Time Multi-Stream Monitoring system for early detection of water quality deterioration by monitoring key parameters simultaneously, using the Brisbane River data for baseline establishment and testing detection performance across multiple control limit configurations.
Department: Data Science and Analytics
Supervisor: Dr. Austin Brown
|
Second Place
Abstract: This work presents AutoTrader-AgentEdge, a human-in-loop trading system that positions AI agents as collaborative partners rather than autonomous replacements. We demonstrate that multi-indicator consensus voting combined with human approval achieves superior risk-adjusted returns while maintaining interpretability and control. Core Contribution: A production-ready VoterAgent implementing democratic voting between MACD momentum and RSI extremes, generating transparent trading signals for human evaluation. Unlike black-box automation, our interactive CLI augments trader expertise through interpretable consensus logic. The human retains final decision authority at all critical junctures. Validated Performance: Empirical validation demonstrates multi-indicator voting superiority over single-indicator automation: Sharpe ratio 0.856 vs 0.841, max drawdown -10.10% vs -10.58%, win rate 51.4% vs 31.9%. Extended testing shows 11.2% better relative performance in volatile markets, validating risk management focus. Implementation: Built on the Microsoft AutoGen framework with an extensible multi-agent architecture. SQLite caching achieves 8-10x performance improvement. Interactive CLI enables natural language trade discussion with human approval gates, ensuring trader control while reducing cognitive load. Alpaca broker integration for paper trading. Key Insight: Transparent, interpretable methods build trust and enable effective human-AI collaboration. By prioritizing augmentation over automation, we demonstrate that AI serves traders best as a decision support tool that preserves human judgment while systematically reducing errors. Implications: Contributes to augmented trading research - systems designed to enhance rather than replace human expertise. This work validates that human-in-loop architectures can achieve both superior risk-adjusted returns AND maintained trader control, challenging the "automation-at-all-costs" paradigm prevalent in algorithmic trading.
Department: Information Technology
Supervisor: Dr. Ying Xie
| |
Third Place
Abstract: In Frequency Division Duplex (FDD) 5G networks, downlink channel state information (CSI) must be estimated at the user equipment (UE) and fed back to the base station, a process that requires frequent CSI-RS transmission and uplink feedback, resulting in high overhead and energy consumption. This research proposes a novel base-station–centric framework that predicts the downlink channel matrix directly at the gNB, eliminating the need for continuous CSI-RS–based estimation at the UE. By leveraging uplink channel observations, geometric environment features, and learned mappings between uplink and downlink channel relationships, our model reconstructs the downlink MIMO channel with high fidelity. The system integrates ray-tracing-based dataset generated via NVIDIA Sionna, combined with graph attention networks and transformer encoders to capture spatial and temporal channel dependencies. We infer the downlink channel without per-slot CSI-RS transmission based in environment geometry and uplink transmission signals, significantly reducing signaling overhead while maintaining beamforming performance. This work enables a shift toward AI-assisted FDD systems where proactive channel prediction replaces periodic downlink probing, contributing to greener and more efficient 5G/6G networks.
Department: Computer Science
Supervisor: Dr. Ahyoung Lee
Audience Favorite Presenter
Abstract: Classical machine learning methods such as CNNs, SVMs, PCA, Logistic Regression, and Random Forests, have achieved strong performance across fields from computer vision to drug discovery. However, these models face scalability limits when trained on large or high-dimensional datasets. Quantum computing introduces quantum mechanics like superposition, interference, and entanglement - enabling quantum kernels, quantum feature maps, and hybrid quantum-classical architectures that may reduce computational cost or enhance data representation. This project implements classical versions of these algorithms alongside their quantum counterparts to evaluate differences in accuracy and efficiency. By comparing performance across diverse scientific and engineering datasets, the study assesses whether quantum-enhanced methods can match or surpass classical baselines under practical constraints.
Department: Computer Science
Supervisor: Prof. Sharon Perry
KSU Supervisor: Dr. Yong Shi
|