Development and Characterization of Synthetic Riboswitches for Small Molecule Metabolic Biomarkers

Type: Undergraduate
Author(s): Kadie Nguyen Biology
Advisor(s): Youngha Ryu Chemistry & Biochemistry
Location: FirstFloor, Table 11, Position 1, 1:45-3:45

This research aims to develop and characterize synthetic riboswitches for creatinine and trimethylamine N-oxide (TMAO), metabolic biomarkers for kidney and cardiovascular dysfunctions. Riboswitches are structured RNA elements located in the 5’-untranslated regions (UTRs) of bacterial mRNAs that regulate downstream gene expression through ligand-induced conformational changes with high affinity and selectivity. To select for the synthetic riboswitches specific to creatinine, the glycine riboswitch library was subjected to a dual genetic selection. In the positive selection, the riboswitches that bind to creatinine or any endogenous molecules will produce the CAT-UPP fusion protein, allowing the host cells to survive in the presence of chloramphenicol. The negative selection is carried out in media containing 5-fluorouracil (5-FU) in the absence of creatinine. Any riboswitches activated by endogenous ligands will die in the presence of 5-FU. The surviving cells should contain the riboswitches that are activated only by creatinine. After several repeated selection steps, including increased concentrations of chloramphenicol and 5-FU, no glycine riboswitch variants were identified to show chloramphenicol resistance in the presence of creatinine. We will continue the project with different riboswitch libraries. We identified a synthetic riboswitch to TMAO, a riboswitch that was derived from the genetic selection of the theophylline riboswitch library that clearly shows chloramphenicol resistance only in the presence of TMAO. We will further test this TMAO riboswitch by colorimetric or fluorescence assays using β-galactosidase and green fluorescence protein, respectively, in the presence of varying concentrations of TMAO.

Self-Propagating High-Temperature Synthesis of Silicon using Plant-based Silica

Type: Undergraduate
Author(s): Iris Nguyen Chemistry & Biochemistry
Advisor(s): Jeffery Coffer Chemistry & Biochemistry
Location: SecondFloor, Table 6, Position 1, 1:45-3:45

Sustainable synthetic approaches to drug delivery carriers such as porous silicon are becoming increasingly important for biomedical applications such as drug delivery, where extreme electronic-grade purity is not required, even though silicon remains a critical material in electronics and energy technologies. This work develops a green, self-propagating high-temperature synthesis (SHS) approach to produce high-surface-area porous silicon using plant-derived silicon dioxide (SiO₂) as the precursor, magnesium (Mg) as the reductant, and sodium chloride (NaCl) as a thermal moderator. The exothermic magnesiothermic reaction is initiated using a controlled electrical input of less than (or equal to) 9V, enabling silicon formation while significantly reducing external energy requirements compared to conventional high-temperature silicon production methods.

In practice, Mg and SiO₂ reactants are exposed to a finite voltage for approximately 10–15 minutes to allow the SHS reaction to propagate. After synthesis, the crude product is purified by dissolving reaction byproducts in concentrated hydrochloric acid, leaving behind porous silicon. X-ray powder diffraction (XRD) is used to evaluate crystallinity and phase composition. While XRD analysis confirms the formation of silicon, persistent crystalline silica peaks indicate incomplete reduction and phase coexistence that currently limits effective separation. Ongoing work focuses on optimizing reaction conditions and refining reaction kinetics to improve phase selectivity and identify optimal synthesis parameters. Despite these challenges, the low-energy synthesis strategy and use of accessible raw materials highlight the potential of SHS-derived porous silicon as a scalable and sustainable platform for future drug delivery applications, particularly in resource-limited settings.

Breaking the Keap: Unlocking Nrf2 for Neuroprotection in Alzheimer’s Disease.

Type: Graduate
Author(s): Hannah Pyle Chemistry & Biochemistry Nitish Kumar Chemistry & Biochemistry David Mingle Chemistry & Biochemistry
Advisor(s): Kayla Green Chemistry & Biochemistry
Location: Third Floor, Table 12, Position 2, 1:45-3:45

Oxidative stress plays a significant role in the progression of Alzheimer’s disease, making cellular antioxidant pathways attractive therapeutic targets. The Keap1–Nrf2 signaling pathway regulates the cellular response to oxidative stress, and inhibition of the Keap1 protein can activate Nrf2, promoting neuroprotective antioxidant responses. In this study, a series of quinoline-modified macrocyclic compounds were designed and synthesized to evaluate their potential as Keap1 inhibitors.
Computational and experimental approaches were employed to investigate the interaction of these compounds with the Keap1 protein. In-silico studies were conducted to analyze the binding affinity of the synthesized compounds using molecular docking, molecular dynamics simulations, and machine learning–based prediction of IC₅₀ values. These analyses provided insight into the stability of the ligand–protein complexes and the structural features that influence binding interactions.
The computational results indicate that compounds containing polar substitutions on the upper synthon exhibit stronger binding affinity and form more stable complexes with the Keap1 protein. Additionally, modification of the macrocyclic scaffold with quinoline substitution on the side nitrogen was found to enhance interactions with the protein binding pocket, suggesting a favorable structural motif for Keap1 inhibition.
Together, these findings provide insight into structure–activity relationships for this class of compounds and highlight promising molecular features for the development of Keap1 inhibitors as potential therapeutic leads for Alzheimer’s disease.

Fighting Cancer with Macrocycles: New Molecular Shapes Provide New Treatment Strategies

Type: Graduate
Author(s): Andrea Rangel Chemistry & Biochemistry
Advisor(s): Eric Simanek Chemistry & Biochemistry
Location: SecondFloor, Table 9, Position 3, 11:30-1:30

Chemotherapy relies on two therapeutic paradigms. The classic approach, most often used, employs small molecules to specifically target enzyme active sites, as represented by the new generation of kinase inhibitors. A secondary approach relies on interfering with protein-protein interactions thus requiring the use of larger compounds. While this latter strategy is garnering the attention of the pharmaceutical community, the rules for the design of these larger molecules, which are often cyclic, are not understood. The compact shape of small molecules leads to predictable behaviors including oral availability and cell uptake. For larger molecules that adopt multiple shapes, understanding the factors that control their shape and dynamic motion provides opportunities to predict similar behaviors that are critical for rational drug design. Here, the synthesis and characterization of a library of large, cyclic molecules (macrocycles) is described. The macrocycles of interest result from the dimerization of monomers. A total of 50 monomers containing different drug-like groups were synthesized. Reaction of a single monomer yields a homodimer, while combination of two different monomers leads to a 1:1:2 mixture of homodimers and a heterodimer. These combinations ultimately lead to a library of 1,275 different compounds. Liquid chromatography-mass spectrometry confirms that >99.9% of the reactions were successful. To investigate the biological activity of these compounds, we have provided this library to high throughput drug-screening facilities at Vanderbilt University and Scripps Florida. Of the several compounds created, macrocycles containing hydroxylamine groups are of special interest for two reasons. First, these molecules are similar to Hydrea, a widely-used, FDA-approved cancer drug. Second, unlike most macrocycles, both the shape and dynamics of these molecules are well understood so critical parameters such as oral availability and membrane transit can be predicted.

Targeting Oxidative Stress in Alzheimer’s Disease through Mechanistic Design of SOD-Mimicking Copper and Manganese Macrocycles

Type: Undergraduate
Author(s): Mark Sayegh Chemistry & Biochemistry Katie Smith Chemistry & Biochemistry
Advisor(s): Kayla Green Chemistry & Biochemistry
Location: SecondFloor, Table 4, Position 3, 11:30-1:30

Reactive oxygen species (ROS) are byproducts of normal cellular metabolism and play important roles in cell signaling and immune defense. However, when their production exceeds the cell’s antioxidant capacity, ROS accumulation leads to oxidative stress, damaging proteins, lipids, and DNA. In the brain, this oxidative imbalance has been closely linked to the development and progression of neurodegenerative diseases like Alzheimer’s. Under normal conditions, superoxide dismutase (SOD) enzymes play a key role in protecting cells by breaking down harmful superoxide radicals. Yet, reduced SOD activity and impaired regulation have been consistently observed in patients with neurodegeneration, including Alzheimer’s disease. Small-molecule mimics of SOD, therefore, represent a promising therapeutic approach. In this study, we evaluate an expanded library of tetra-aza macrocyclic ligands chelating either copper or manganese metals. Mechanistic analysis reveals how structural modifications to the macrocyclic ring, particularly R-group substitutions that alter steric environment and electronic properties, directly influence catalytic reactivity and stability. Evaluation of Cu- and Mn-based complexes highlights distinct trends in activity and identifies structural motifs that enhance SOD-like function. These findings provide design principles for developing antioxidant therapeutics targeting oxidative stress.

Superfrog Science: Experiments for at Home and in the Classroom

Type: Undergraduate
Author(s): Samantha Shah Chemistry & Biochemistry Peyton Green Chemistry & Biochemistry
Advisor(s): Kayla Green Chemistry & Biochemistry
Location: SecondFloor, Table 7, Position 3, 1:45-3:45

“Superfrog Science: Experiments for at Home and in the Classroom” is a creative endeavor that encourages young scientists to get curious about science and help learn a variety of chemistry concepts. This book is a visual representation of the importance of exploring the bounds of creativity in science. Join Superfrog as he goes on a learning adventure conducting science experiments and using the scientific method to deepen his knowledge about chemistry, making it easy to learn by conducting each experiment with easy-to-follow comic panels. This book strives to make science experiments accessible, affordable, and fun. It is perfect for encouraging hands-on learning, with in-depth explanations of the “how” and “why” of these experiments. Superfrog asks discussion questions and provides variations of the experiments to get his students to think about cause-and-effect and variable manipulation when it comes to the scientific process, as well as encourage collaboration with his peers. The educational activities featured are made for scientific discovery inside and outside the classroom. Make chemistry fun and easy with Superfrog!

Engineering Sarcosine and Uric Acid Aptamers via Riboswitch-Based Dual Selection for Biomarker Detection

Type: Undergraduate
Author(s): Jeremiah Tran Chemistry & Biochemistry
Advisor(s): Youngha Ryu Chemistry & Biochemistry
Location: SecondFloor, Table 5, Position 3, 11:30-1:30

Riboswitches are structured RNA elements that regulate gene expression through ligand-induced conformational changes and provide a platform for engineering cell-based biosensors. By coupling aptamers to reporter genes, synthetic riboswitches enable small-molecule–dependent detection of clinically relevant metabolites. This study focuses on sarcosine, associated with prostate cancer progression, and urate, linked to gout. Two sarcosine-responsive candidates were evaluated in E. coli using β-galactosidase and GFP reporter systems. Although construct integrity was confirmed, neither candidate demonstrated ligand-dependent activation in CDR or minimal media, suggesting insufficient regulatory activity under tested conditions. In parallel, a urate-responsive riboswitch library underwent dual selection with chloramphenicol resistance for positive selection and 5-fluorouracil counterselection for negative selection. After multiple selection rounds and screening of 192 colonies, no urate-specific variants were identified. Increasing chloramphenicol concentration to strengthen positive selection similarly yielded no hits. Future work will focus on further increasing both positive and negative selection intensity to enhance enrichment of functional variants and improve development of RNA-based biosensors for accessible metabolite detection. Additionally, future efforts will explore the adenine riboswitch library as a potential platform for developing novel biomarker detection systems.

Computational Analysis of Nitric Oxide Dioxygenase Biomimicry with Non-heme Small Molecules

Type: Graduate
Author(s): Caden-Jack Walls Chemistry & Biochemistry Kayla N. Green Chemistry & Biochemistry Benjamin Janesko Chemistry & Biochemistry
Advisor(s): Kayla Green Chemistry & Biochemistry
Location: SecondFloor, Table 6, Position 1, 11:30-1:30

Nitric oxide (NO) is a gaseous free-radical 2° messenger with a physiological half-life of 3-5 seconds. Overexpression of the cytoprotective NO can lead to high concentrations of cytotoxic peroxynitrite (OONO^-), causing nitroxidative stress. Studies have shown that nitroxidative stress can be implicated as an etiology of several inflammatory diseases, such as Alzheimer’s Disease (AD) or Parkinson’s Disease (PD). A solution to counter nitroxidative stress is the biomimicry of the enzyme Nitric Oxide Dioxygenase (NOD). The enzymic activity of NOD relies on a heme active site, where excess NO is scavenged to produce nitrate (NO_3^-), a less potent oxidant. Several groups have successfully mimicked this activity; however, it has been restricted to water-insoluble, large molecules (porphyrin rings). While other antioxidant enzymes such as Superoxide Dismutase and Catalase have been successfully mimicked with water-soluble, metal-centered, non-heme scaffolds, to date, there have been no reports of water-soluble non-heme mimics of NOD activity. It is the Green Group’s goal to explore the possibility of developing a molecule capable of NOD enzymic activity. Therefore, theoretical feasibility of this reaction was explored using Density Functional Theory (DFT) as well as Conformer-Rotamer Ensemble Sampling Tool (CREST). Current data shows that based on an energy screening of several simple-to-complex tetra-aza small molecules, the reaction is successful both in gas phase and in water (implicit and explicit solvation). Additionally, computational intermediate spin states have, so far, matched those reported experimentally. Energy diagrams were then proposed based on the most stable ground state energies of structural intermediates. This data provides, for the first time, a new perspective on the possibility of the successful biomimicry of NOD with non-heme, water-soluble, tetra-aza small molecules.

Interactive Querying and Visualization of Solar Events

Type: Undergraduate
Author(s): Thu My Banh Computer Science Robin Chataut Computer Science
Advisor(s): Pandey Chetraj Computer Science
Location: Basement, Table 2, Position 1, 1:45-3:45

Interactive Querying and Visualization of Solar Events

Author: Thu My Banh, Cathy Nguyen, Pandey Chetraj

Access to structured solar flare event data is essential for space weather (SWx) research, operational analysis, and machine learning applications. While the solar flare event archive maintained by the Lockheed Martin Solar and Astrophysics Laboratory (LMSAL) provides a widely used curated record of flare activity, the archive is primarily accessible through static web interfaces rather than a programmable query system. This makes automated filtering, dataset generation, and large-scale analysis difficult for researchers. To address this limitation, we developed a full-stack web application that provides programmatic access to LMSAL solar flare event records through a queryable API. A Python-based data ingestion pipeline retrieves and deduplicates event information from LMSAL’s rolling snapshot archive and stores it in a structured format. A FastAPI backend exposes endpoints that allow users to filter events by date range and GOES flare classification, enabling rapid dataset generation for analysis workflows. The frontend, implemented in React, allows users to query the event catalog, visualize results in a structured table, and export filtered datasets as CSV or JSON files. To improve data reliability and context, the system cross-references LMSAL event records with NOAA solar flare catalogs, allowing users to compare event metadata across independent data sources. Additionally, the application integrates with the Helioviewer API to display solar imagery corresponding to each event, with derived heliographic positions overlaid onto the solar disk to provide spatial context. The resulting system provides a lightweight platform for exploring, querying, and exporting solar flare event data, lowering the barrier to accessing operational flare records and facilitating dataset generation for space weather analysis and predictive modeling.

Capability Boundaries of LLM-Assisted Cyberattacks: An Empirical Evaluation

Type: Undergraduate
Author(s): Gabriella Campos Computer Science Jayapradeep Jayaraman Srinivas Computer Science Tam Nguyen Computer Science Riley Phan Computer Science Rahul Shrestha Computer Science
Advisor(s): Robin Chataut Computer Science
Location: SecondFloor, Table 12, Position 1, 1:45-3:45

Large language models (LLMs) are increasingly framed as force multipliers for cyberattacks, yet most existing evaluations focus on isolated artifact generation rather than the construction and execution of full offensive workflows. This paper presents a controlled empirical study of LLM-assisted cyberattack construction across multiple representative attack classes, including automated SQL injection exploitation, spyware assembly, reverse shell establishment, and denial-of-service traffic generation. We evaluate several contemporary models—including ChatGPT-4o, ChatGPT-5.2, ChatGPT-5.1-instant, Claude Sonnet 4.6, and Gemini 3—within fully sandboxed virtualized environments, treating each model strictly as an advisory system embedded within a human-driven workflow.

Our experimental design decomposes attacks into staged operational workflows encompassing reconnaissance, payload generation, system integration, troubleshooting, and persistence. This structure enables systematic analysis of where automation succeeds or fails during real execution rather than relying on single-shot demonstrations. Across scenarios, LLMs consistently reduce effort for localized technical tasks such as command syntax recall, tool configuration, payload scaffolding, and procedural troubleshooting. However, reliable end-to-end attack execution remains limited. SQL injection automation succeeds primarily when established tools encapsulate complex orchestration, while more complex scenarios such as spyware assembly fail at system-level integration, environment-specific dependency resolution, and evasion of host defenses.

Across models and attack classes, automation consistently breaks at environment-dependent boundaries requiring global reasoning, state awareness, and cross-stage workflow coordination. These findings suggest that contemporary LLMs do not autonomously execute cyberattacks but instead function as workflow accelerators that lower the expertise threshold required to operationalize existing offensive techniques. This capability-boundary perspective provides a more realistic foundation for threat modeling, defensive planning, and future evaluation of AI-assisted cybersecurity risks.

DiseaseQuest

Type: Undergraduate
Author(s): Carlota Castelltort Pinto Computer Science Alexander Canales Computer Science Long Dau Computer Science Chris Musselman Computer Science Dylan Noall Computer Science Rahul Shrestha Computer Science Kavish Soningra Computer Science
Advisor(s): Bingyang Wei Computer Science
Location: SecondFloor, Table 6, Position 3, 11:30-1:30

Medical students lack effective tools for developing clinical reasoning, as most resources emphasize memorization rather than decision-making. DiseaseQuest is an AI-powered, gamified platform that addresses this gap through realistic patient simulations and decision-based scenarios. It enables students to work through complete clinical cases using interactive, patient-centered dialogue. Supported by a multi-agent framework, the platform provides adaptive guidance, diagnostic feedback, and personalized evaluations. By promoting active learning and problem-solving, DiseaseQuest offers a transformative approach that replaces passive study with immersive, hands-on practice, helping students strengthen diagnostic thinking and better prepare for real-world clinical decision-making.

NutrimindAI

Type: Undergraduate
Author(s): Mayra Coronilla Computer Science Sujit Bhandari Computer Science Samiksha Gupta Computer Science Michelle Jimenez Computer Science Kim Nguyen Computer Science Keilah Scott Computer Science Nibesh Yadav Computer Science
Advisor(s): Xi Fitzgerald Computer Science
Location: Third Floor, Table 18, Position 1, 11:30-1:30

As obesity continues to rise in the United States, bariatric surgery has become as increasingly common medical intervention to support significant and sustained weight loss. However, the procedure presents challenges, as patients must adopt strict dietary guidelines, develop consistent meal tracking habits, and maintain long-term lifestyle changes. Existing weight-loss applications fail to address the unique nutritional requirements of bariatric patients, which include surgery-specific restrictions, medical conditions, personal preference in food, and individualized lifestyle factors. Along with that, they lack integrated long-term monitoring tools that allow healthcare providers to effectively track patient progress and adherence after surgery. This senior design project presents a prototype mobile application developed from scratch to support patients throughout the bariatric journey. The application integrates AI-driven personalization to generate tailored daily nutritional guidance, adapt to individual health data, and provide meal tracking support. In addition, the platform centralizes patient data for healthcare providers, improving long-term monitoring, increasing tracking accuracy, and reducing manual workload. By combining personalized patient support with provider-facing analytics, this solution aims to enhance postoperative adherence and improve long-term surgical outcomes.

Brazos Safety Systems - Weather System

Type: Undergraduate
Author(s): Robert Hannaford Computer Science Iyed Acheche Computer Science Oscar Arenas Computer Science Nagendra Chaudhary Computer Science Evan Eissler Computer Science Tucker Rinaldo Computer Science Sumalee Rodolph Computer Science
Advisor(s): Ed Ipser Computer Science
Location: Basement, Table 10, Position 2, 11:30-1:30

Understanding weather conditions during flight operations can help explain incidents and reduce risky behavior. The Brazos Safety Systems Weather Application integrates aviation weather data sources, including METAR reports and radar imagery, to visualize conditions around airports and during historical flights. Users can upload flight records and review the associated weather conditions through the application. By presenting aviation weather data in a centralized and accessible format, the application supports post-flight analysis and helps identify weather-related factors connected to flight incidents. The goal is to provide insights that improve understanding of past flight conditions and help prevent similar issues in future aviation operations.

Operational Verification of Deep Learning–Based Solar Flare Forecasts

Type: Undergraduate
Author(s): Son Hoang Computer Science Robin Chataut Computer Science Chetraj Pandey Computer Science
Advisor(s): Chetraj Pandey Computer Science
Location: Basement, Table 11, Position 2, 1:45-3:45

Solar flares are among the most significant drivers of space-weather disturbances, motivating ongoing efforts to develop reliable forecasting methods from solar observations. The Solar Dynamics Observatory continuously produces high-resolution full-disk solar imagery used for monitoring solar activity. These observations have enabled substantial progress in machine learning–based flare prediction; however, most models remain confined to research studies, with limited deployment in operational systems that support continuous forecasting and systematic performance validation. This work presents a lightweight operational framework for near-real-time solar flare forecasting built around machine learning models proposed in the literature. The system retrieves full-disk solar imagery from the Helioviewer API, performs automated preprocessing, and generates predictions using a convolutional neural network–based forecasting model. Predictions and corresponding observations are stored to enable periodic forecast verification using standard performance metrics, allowing model performance to be monitored over time and potential prediction drift to be identified. The framework is implemented as an interactive application using Streamlit, providing an integrated interface for automated data ingestion, near-real-time inference, and ongoing model evaluation. The proposed system enables continuous monitoring of solar flare forecasts while providing a practical framework for tracking model performance and detecting prediction drift in operational settings.

AI-Driven Adaptive Tutoring: A Multi-Agent System for Structured, Multimedia-Enhanced Education

Type: Undergraduate
Author(s): Jayapradeep Jayaraman Srinivas Computer Science Gabriella Campos Computer Science Robin chataut Computer Science Nagendra Chaudhary Computer Science Riley Phan Computer Science
Advisor(s): Robin Chataut Computer Science
Location: Basement, Table 7, Position 1, 1:45-3:45

We present the AI-Driven Adaptive Tutoring (AIAT) framework, a modular multi-agent system that generates structured, retrieval-grounded, and multimedia-enhanced courses. AIAT targets a common gap in AI in Education: large language models (LLMs) can produce fluent explanations, but they often lack pedagogical structure, factual grounding, and multimodal integration. To address this, AIAT uses a three-stage pipeline. First, a blueprint agent creates a course outline with learning objectives and topic dependencies using schema-validated structured outputs. Second, a chapter-expansion agent instantiates atomic topics with formative questions and summaries in JSON mode. Third, an enrichment agent generates topic-level explanations, visualization specifications, and triggers for narrated video production. Retrieval-augmented generation (RAG) combines a MongoDB Atlas Vector Search backend for course materials and a Pinecone pipeline for PDF-derived knowledge, grounding explanations in external content. A Next.js frontend streams responses and assembles text, diagrams, and videos into a unified learner experience.

The design is explicitly guided by mastery learning, cognitive load theory, and the Cognitive Theory of Multimedia Learning, with principles such as atomic topics, anti-fluff constraints, and visual-verbal alignment encoded in prompts and schemas. We report system-level metrics (e.g., latency by component) and operational reliability, and we outline a concrete evaluation plan, including pre/post-learning assessments, expert rubric-based accuracy checks, and subjective cognitive load measures. We also discuss the equity and accessibility implications of relying on commercial APIs and propose mitigation strategies (e.g., caching, partial use of lightweight models, and instructor-in-the-loop authoring). The contribution of this work is a reproducible architecture that connects multi-agent orchestration, RAG, and multimodal rendering to pedagogical theory, along with an evaluation roadmap that explicitly addresses the current lack of large-scale human studies.

HelioIndex: A Python Package for Reproducible Construction of Machine-Learning Datasets from Solar Observations and Event Catalogs

Type: Undergraduate
Author(s): Balaji Kannan Computer Science Robin Chataut Computer Science
Advisor(s): Chetraj Pandey Computer Science
Location: SecondFloor, Table 10, Position 1, 1:45-3:45

Space Weather Forecasting relies on large volumes of time-stamped solar observations paired with event catalogs describing flare occurrence and intensity. While these datasets are widely available, preparing them for machine learning remains a substantial and often overlooked challenge. Researchers must convert irregular observation streams into consistent temporal samples, construct observation and prediction windows, align events with observations, manage missing data and cadence inconsistencies, and ensure that training and evaluation splits avoid temporal or regional data leakage. These preprocessing steps are typically implemented in ad-hoc scripts that are difficult to reproduce, extend, or compare across studies. We propose an open-source Python library that standardizes the construction of machine-learning-ready datasets for solar event forecasting. The library ingests user-provided observation tables (e.g., SDO image timestamps and file paths) and event catalogs (e.g., GOES flare lists) and automatically generates indexed training samples suitable for PyTorch datasets and data loaders. Users can define flexible observation windows ranging from single time points to multi-frame temporal sequences, specify prediction horizons, and configure event-labeling rules. The framework also provides mechanisms for handling missing observations, irregular cadences, and explicit representation of temporal gaps. To support rigorous experimental design, the library includes reproducible dataset partitioning strategies such as chronological and tri-monthly splits, as well as optional active-region-aware grouping based on NOAA region catalogs. These features allow researchers to build both full-disk and active-region-based forecasting datasets while minimizing common sources of information leakage. Although the initial implementation focuses on solar flare prediction, the framework is designed to be extensible to other space weather phenomena including coronal mass ejections (CMEs) and solar energetic particle (SEP) events. By formalizing the transformation from raw observational records and event lists into reproducible machine learning datasets, this tool reduces the overhead of data preparation and promotes more transparent, comparable, and scalable space weather forecasting research.

A Cross-Model Validation Framework for Solar Flare Forecasting Using NASA CCMC Flare Scoreboard Predictions

Type: Undergraduate
Author(s): Kritika Karanjit Computer Science Robin Chataut Computer Science Chetraj Pandey Computer Science
Advisor(s): Chetraj Pandey Computer Science
Location: Basement, Table 5, Position 1, 11:30-1:30

Solar flares are significant space weather phenomena that can impact satellites, communication systems, and many technological infrastructures, rendering accurate flare forecasting a crucial objective in heliophysics study.The NASA Community Coordinated Modeling Center (CCMC) Flare Scoreboard collects predictions from multiple solar flare forecasting models developed by different research groups. While this resource provides a useful platform for comparing different forecasting approaches, systematic validation of these models remains challenging because predictions are reported in different formats and are not easily comparable across models. In this work, we develop an automated framework to collect and organize flare forecasts from several models available in the CCMC Flare Scoreboard and convert them into a consistent dataset that allows direct comparison between models. The processed dataset includes predictions across multiple years and forecast windows. To evaluate model performance, we compare the predicted flare probabilities with observed flare events reported in the SolarSoft (SSW) Latest Events archive. By aligning forecast windows with actual flare occurrences, we establish a consistent approach for validating model predictions. This approach facilitates a systematic comparison of forecasting behaviour among various models and assists in identifying those that exhibit superior or inferior predicted ability.The resulting pipeline provides a reproducible way to analyze solar flare forecasting systems and supports future efforts to improve the reliability of space weather prediction methods.

Integrating Full-Disk and Active-Region Models for Reliable Forecasting of Solar Flares

Type: Undergraduate
Author(s): Duc Le Computer Science Robin Chataut Computer Science Chetraj Pandey Computer Science
Advisor(s): Chetraj Pandey Computer Science
Location: Third Floor, Table 12, Position 1, 11:30-1:30

Solar flares are major drivers of space-weather disturbances and can disrupt satellites, communication systems, and navigation infrastructure. Recent deep learning approaches have demonstrated promising performance for solar flare forecasting, yet many models operate either on full-disk solar observations or on isolated active-region patches. This separation limits their ability to combine global solar context with localized magnetic structure and can affect the reliability of predictions. In addition, full-disk models often provide limited information about which regions drive their forecasts. This study presents a two-stage deep learning framework that integrates full-disk and active-region–level analysis within a unified flare forecasting pipeline. The system first performs full-disk inference using a convolutional neural network trained on solar magnetograms to estimate the global probability of flare occurrence. Attribution-based explanations are then generated to identify regions that most strongly influence the model prediction. These regions are mapped back to the solar disk and converted into candidate active-region patches, accounting for solar rotation and spatial alignment. The resulting patches are subsequently analyzed using a dedicated active-region forecasting model trained on SDO HMI SHARP data to produce localized flare probabilities. By integrating global context with targeted active-region analysis, the proposed framework combines two complementary forecasting models into an end-to-end prediction system. The resulting pipeline provides both full-disk and region-level flare probabilities, improving interpretability while enhancing the reliability of flare forecasts through focused secondary analysis of the most relevant solar regions.

Using AlphaFold2 to Identify Novel Drug Targets Against Cryptococcus

Type: Undergraduate
Author(s): Francisco Lugo Gonzales Computer Science
Advisor(s): Natalia Castro Lopez Biology

Building a Unified Interface for Solar Event Data Retrieval from SWPC Archives

Type: Undergraduate
Author(s): Cathy Nguyen Computer Science Thu My Banh Computer Science
Advisor(s): Chetraj Pandey Computer Science
Location: Third Floor, Table 10, Position 1, 1:45-3:45

Solar event archives from NOAA Space Weather Prediction Center (SWPC) contain observations of solar phenomena such X-ray flares (XRA), optical flares (FLA), disappearing solar filament (DSF), radio bursts (RSP), and other solar events. However, these data are currently stored across multiple sources and incompatible formats. As a result, this makes event retrieval, cross-comparison, and large-scale analysis complicated. In this study, we introduce a computational framework to extract and standardize solar event data from SPWC archives into a unified structure. Our approach automates parsing event reports, extracts key features such as event classification and timing, and organizes them to convert records into a consistent format across datasets. By reducing differences in how event records are stored and represented, this framework can enhance the usability of the solar records. The ultimate goal is to support the development of a tool supporting easier and faster access to solar event records based on user-selected criteria such as event type or time range. This standardization aims to improve data accessibility, providing a foundation for further space weather research.

TAED: A Trust-Aware Explainable Defense for Phishing Detection Under Adversarial Manipulation

Type: Undergraduate
Author(s): Tam Nguyen Computer Science Robin Chataut Computer Science
Advisor(s): Robin Chataut Computer Science
Location: Basement, Table 4, Position 3, 1:45-3:45

Machine learning-based phishing detection systems increasingly rely on high-confidence predictions from deep neural models, yet confidence alone provides limited assurance of reliability in adversarial environments. Small, semantics-preserving manipulations such as homoglyph substitution and paraphrasing can induce confident misclassifications while remaining indistinguishable to human recipients, exposing a critical vulnerability in modern email security pipelines. We present TAED, a Trust-Aware Explainable Defense system that explicitly evaluates prediction trustworthiness rather than relying solely on opaque confidence scores. TAED computes a trust score by integrating model confidence with explanation fidelity, which measures alignment between model reasoning and known phishing indicators, and explanation stability, which quantifies sensitivity to minor input perturbations. We evaluate TAED alongside a diverse set of statistical and neural phishing detectors using a realistic adversarial dataset constructed through multiple evasion strategies. Our results reveal a systematic confidence–robustness paradox in which complex Transformer-based models exhibit strong clean-data performance but substantial brittleness under adversarial manipulation, while simpler feature-based models demonstrate greater resilience. By leveraging explanation-derived trust signals and selective escalation within a hybrid detection pipeline, TAED identifies unreliable high-confidence predictions and improves robustness against adversarial evasion. These findings demonstrate that explainability can be operationalized as a practical security mechanism for assessing model reliability in adversarial phishing detection systems.

Viral Agent Based Model Interface

Type: Undergraduate
Author(s): Ellion Norwood Computer Science Hebert Alvarez Computer Science Gabby Campos Computer Science Aqil Dhanani Computer Science Derek Le Computer Science Bereket Mezgebu Computer Science Stefan Saba Computer Science
Advisor(s): Xi Fitzgerald Computer Science
Location: FirstFloor, Table 13, Position 1, 11:30-1:30

Agent-based models (ABMs) are widely used in computational biology to simulate complex processes such as infectious disease transmission. However, many research-grade models are implemented primarily as backend systems and lack graphical interfaces that allow researchers to efficiently configure simulations and interpret outputs. In collaboration with the Biophysics Department, this project focused on the development of a graphical user interface (GUI) for an existing viral agent-based simulation platform previously implemented without an interactive frontend.
The implemented interface integrates with the existing backend simulation environment deployed on laboratory systems, enabling structured parameter configuration, simulation execution, and visualization of model outputs. Development focused on frontend architecture, parameter validation mechanisms, backend connectivity, and data visualization components for simulation result analysis. Additional work included interface refactoring and codebase cleanup to improve maintainability and usability.
The resulting system provides a structured workflow for configuring and executing simulations while preventing invalid parameter configurations through input validation. By extending the existing modeling framework with a robust graphical interface and visualization capabilities, this work improves accessibility and operational efficiency for researchers conducting computational epidemiology experiments within the laboratory environment.

Machine Performance Check Plus — Quality Assurance for Varian TrueBeam Systems

Type: Undergraduate
Author(s): Brae Ogle Computer Science Tristan Gonzales Computer Science Alex Lee Computer Science Alexandre Morales Computer Science Sameep Shah Computer Science Madhavam Shahi Computer Science
Advisor(s): Bingyang Wei Computer Science
Location: FirstFloor, Table 10, Position 1, 11:30-1:30

Machine Performance Check Plus (MPC+) is a software platform designed to improve quality assurance workflows for Varian TrueBeam linear accelerators used in radiation therapy. The system automatically collects and processes Machine Performance Check (MPC) data generated by clinical machines, including imaging files and measurement results, and converts them into structured, analyzable information. The platform provides a web-based dashboard that allows medical physicists and clinical staff to review machine performance metrics, visualize trends, and quickly identify values that fall outside acceptable tolerances. MPC+ also supports digital sign-off workflow and audit trails to ensure compliance with regulatory and clinical standards. By consolidating data from multiple machines and clinics into a single interface, the system reduces the time required for daily QA review while improving reliability and traceability. Overall, the project aims to make the quality assurance process more efficient, data-driven, and scalable for radiation oncology clinics operating Varian TrueBeam systems.

MAVENCODE - Senior Design Project

Type: Undergraduate
Author(s): Emmanuel Oyawoye Computer Science Zaid Alaqqad Computer Science Hayden Brigham Computer Science Michael Dugle Computer Science Tanner Hendrix Computer Science Arscene Rubayita Computer Science Merci Yohana Computer Science
Advisor(s): Shelly Fitzgerald Computer Science
Location: SecondFloor, Table 8, Position 1, 11:30-1:30

This senior design project centers on VANTAGE (Visual Autonomous Navigation and Task-driven Agentic Ground-to-air Engine), an AI-driven drone operations platform developed with MavenCode, a leading AI/ML solutions provider in the Dallas–Fort Worth area. VANTAGE enables users to command drones through natural language while integrating real-time perception tools such as speech-to-text, text-to-speech, object detection, semantic segmentation, and vision-language reasoning. The system combines a FastAPI backend, agent-based tool orchestration, and a web dashboard that supports both mission control and direct testing of AI tools through uploaded audio, image, and video inputs. Our team’s work spans full-stack development, model integration, and interface design to deliver a practical, user-centered platform for intelligent aerial autonomy. Together, these components demonstrate an end-to-end AI product approach that aligns with MavenCode’s mission of empowering organizations through training, product development, and consulting.

An Age-Sensitive Benchmark for Safety Disparities and Representational Bias in LLM-Generated Health Advice

Type: Undergraduate
Author(s): Riley Phan Computer Science
Advisor(s): Robin Chataut Computer Science
Location: SecondFloor, Table 4, Position 2, 11:30-1:30

Large language models (LLMs) such as ChatGPT, Claude, Gemini, and Llama are increasingly being deployed as search and decision-support tools for health-related inquiries. As users provide demographic context, including age, to obtain personalized guidance, these systems can differentially adjust tone, directive strength, or safety framework. Although age can be clinically relevant, unintended variation in the generated advice can introduce systematic safety disparities or representational bias. In this study, we analyze outputs from two major LLM families across 10,679 physical and mental health scenarios to examine how explicit age cues, including teen, young adult, middle-aged, and senior, influence the safety and linguistic properties of generated health advice. To quantify these effects, we introduce three task-specific evaluation metrics: Age Differential Safety Bias (ADSB) to measure relative safety shifts under demographic conditioning, Safety Risk Score (SRS) to capture cumulative weighted unsafe recommendations, and Tone Differential Index (TDI) to detect systematic changes in linguistic complexity and formality associated with representational harm. The results indicate that explicit age cues systematically alter the behavior of the model. Demographic conditioning consistently reduces safety quality relative to age-neutral baselines. Middle-aged cohorts exhibit a higher cumulative safety risk in directive responses, whereas senior cohorts demonstrate elevated tone shifts consistent with oversimplification and increased formality. These findings suggest that demographic sensitivity can introduce measurable allocative and representational disparities in healthcare-oriented LLM systems. This work establishes a reproducible audit framework for evaluating demographic safety sensitivity in health-focused LLM deployments and contributes to the development of standardized evaluation protocols for safer and more equitable integration of AI systems in clinical and consumer health environments.