The Office of Research Cyberinfrastructure invites you to attend the Third Annual UCF Research Computing Symposium that will take place on September 23rd, 2026, at the Cape Florida Ballroom, Student Union.
The symposium will provide a forum for UCF faculty, post-docs and students to present their work and exchange ideas in an informal setting, showcasing how they are leveraging advanced computational resources to advance state-of-the-art in their respective fields.
Featuring: a keynote address, invited talks, poster presentations and networking opportunities with researchers and innovators across campus.
Space is limited and registration is required to attend the event.
This event is sponsored by DataDirect Networks (DDN) and Software House International (SHI).
Student Poster Competition
Participate in the Student Poster Competition; showcase your work in high performance computing, AI/ML, quantum computing and cloud computing; and compete for awards! View call for posters for details.
This year’s Student Poster Competition is sponsored by Amazon Web Services (AWS).
Agenda
| Event | Start | End | Presenter(s) |
|---|---|---|---|
| Sign-In & breakfast (Sponsored by DDN and SHI) | 8:15 AM | – | – |
| Opening Remarks | 8:30 AM | 8:45 AM | Dr. Shafaq Chaudhry, Director, Research Technology |
| Keynote Panel: Digital Twins: Challenges and Opportunities | 8:45 AM | 9:45 AM | Moderated by Dr. Shafaq Chaudhry |
| Faculty Invited Talks – Session I | 9:45 AM | 10:45 AM | Moderated by Dr. Fahad Khan |
| Break | 10:45 AM | 11:00 AM | – |
| Faculty Invited Talks – Session II | 11:00 AM | 12:15 PM | Moderated by Benjamin Keene |
| Networking Lunch (Sponsored by DDN and SHI) | 12:15 PM | 1:00 PM | – |
| Invited Talk by AWS | 1:00 PM | 1:30 PM | – |
| Faculty Invited Talks – Session III | 1:30 PM | 2:45 PM | Moderated by Ezequiel Gioia |
| Break | 2:45 PM | 3:00 PM | – |
| Student Poster Session & Judging (Sponsored by AWS) | 3:00 PM | 4:45 PM | – |
| Award presentations and closing remarks | 4:45 PM | 5:00 PM | – |
Keynote Panel: Digital Twins: Challenges and Opportunities
45 min panel discussion and 15 min audience Q&A
Abstract: Digital twins are dynamic virtual representations of physical assets, processes, and environments that are fundamentally reshaping modern engineering, healthcare, smart cities, and industrial manufacturing. By synchronizing continuous real-time telemetry with high-fidelity simulation and advanced artificial intelligence, digital twins unlock unprecedented operational visibility, proactive maintenance, and data-driven decision-making across complex lifecycles.
Despite this transformative promise, wide-scale enterprise adoption introduces formidable technical and organizational challenges. Organizations must navigate fragmented data silos, legacy system interoperability, heightened cybersecurity risks, model fidelity trade-offs, and an evolving landscape that still lacks universal standardization. Furthermore, these cyberinfrastructure systems require extensive computational capabilities for high-frequency, high-fidelity, and bidirectional synchronization between physical and digital environments.
This keynote panel convenes leading industry pioneers, system architects, and researchers to dissect the current state and future horizon of digital twins. Panelists will explore actionable strategies for overcoming integration friction, establishing robust data governance, and maximizing long-term return on investment. Attendees will gain essential insights into bridging the physical-digital divide to build scalable, resilient, and truly adaptive digital twin ecosystems.
Panelists:
Dr. Soheil Sabri, Assistant Professor and the Director of the Urban Digital Twin Lab at UCF
Bio: Dr. Sabri leads pioneering research in urban digital twins, covering areas like real-time geospatial AI (GeoAI) analysis for urban emergency management, Large Language Model (LLM)-based automation for planning and building compliance assessment, and multi-modal spatial intelligence for critical infrastructure design and operation optimization. He significantly influences digital urban planning through his roles in various professional organizations, including co-chairing the Academia and Research Working Group within the Digital Twin Consortium (DTC), the American Society for Photogrammetry and Remote Sensing (ASPRS), and Urban Digital Twin standards at the Open Geospatial Consortium (OGC). He contributes to the Digital Twin workforce development, training, and education, including the Digital Twin Graduate Certification at the School of Modeling, Simulation and Training.
Dr. Sean Mondesire, Assistant Professor at the School of Modeling, Simulation, and Training at UCF
Bio: Dr. Sean Mondesire is an Assistant Professor at the University of Central Florida’s (UCF) School of Modeling, Simulation, and Training (SMST). He is a part of the Knights Digital Twin Initiative and directs the High-performance Artificial Intelligence Laboratory (HAIL). His research specialties are machine learning and big data analytics for real-time systems and autonomous decision-making at scale. His current research focuses on creating real-time digital twins of advanced manufacturing, simulating what-if scenarios in virtual environments, and providing AI insights to chip manufacturers.
Dr. Ghaith Rabadi, Professor and Director of the School of Modeling, Simulation, and Training at UCF
Bio: Professor and Director at the School of Modeling, Simulation, and Training (SMST) at the University of Central Florida (UCF), Orlando, FL. He received his Ph.D. and M.S. in Industrial Engineering from UCF in 1999 and 1996 respectively, and his B.S. in Industrial Engineering from Jordan University in 1992. Prior to joining UCF in 2022, he was a Professor and Graduate Program Director at The Engineering Management and Systems Engineering Department at Old Dominion University, Norfolk, Virginia. His research and teaching interests include Modeling & Simulation, Digital Twins, Planning and Scheduling, Operations Research, Optimization, and Artificial Intelligence in various areas of application including supply chains, transportation, production & manufacturing, disaster response, among others. He has published over 140 peer reviewed journal and conference articles, 3 books, and many abstracts. His research has been funded by various sources including US Department of Education, NASA, NATO, US Army, Department of Homeland Security (DHS), and Virginia Port Authority (VPA), among others. More information can be found at https://ghaithrabadi.com/.
Dr. Minas Pantelidakis, Assistant Professor in the Department of Industrial Engineering and Management Systems at UCF
Bio: Dr. Pantelidakis leads UCF’s Digital Transformation and Cyber-physical Systems (DT & CPS) Laboratory, where his team advances Digital Twin research across Digital Manufacturing, Digital Agriculture, and Distributed Manufacturing through cutting-edge research, interdisciplinary collaboration and education.
Dr. Pantelidakis holds a Ph.D. (2025) and an M.S. (2024) in Industrial and Systems Engineering from Auburn University, as well as an MEng. in Electrical and Computer Engineering from the Technical University of Crete (2020). His academic accomplishments have been recognized with the JT Black Lean Engineering Student of the Year award (2023) and the Walt and Virginia Woltosz Fellowship (2021–2025).
During his doctoral studies, Dr. Pantelidakis was a researcher at the Interdisciplinary Center for Advanced Manufacturing Systems (ICAMS) at Auburn University, where he led innovative research exploring the integration of Digital Twins into manufacturing, supported by funding from the Department of Defense (DoD), the National Institute of Standards and Technology (NIST), and various industry collaborators. Additionally, during his undergraduate studies, he led a precision irrigation initiative in partnership with the Center of Irrigation Technology (CIT) at California State University, Fresno, with funding from the California Department of Food & Agriculture (CDFA). His scholarly contributions include publications in prestigious journals such as the Journal of Manufacturing Systems, the International Journal of Advanced Manufacturing Technology, the International Journal of Lean Six Sigma, and Expert Systems with Applications.
Faculty Invited Talks – Session I
15-minute presentations
Dr. Pegah Khosravi, Enabling Trustworthy Multimodal Medical AI Through Research Computing: The Kidney-VLM Project
Abstract: Developing clinically useful artificial intelligence requires more than innovative algorithms; it also depends on scalable computing, secure data infrastructure, and the ability to integrate large, heterogeneous biomedical datasets. This talk will introduce Kidney-VLM, a multidisciplinary research project developing an explainable vision-language model for kidney cancer diagnosis and characterization. The project will integrate CT imaging, pathology, and clinical information from a large real-world patient cohort to generate clinically interpretable predictions and supporting evidence. I will discuss how high-performance computing, GPU resources, and large-scale storage enable multimodal model development, validation, and reproducible analysis while addressing challenges related to data volume, privacy, model reliability, and clinical translation. Kidney-VLM provides a representative example of how research cyberinfrastructure can help transform complex biomedical data into trustworthy AI systems designed to support physician decision-making and advance precision medicine.
Dr. Thomas Wahl and Dr. Meghana Nagaraj, Global Deep Learning Model for Coastal Storm Surge Reconstruction
Abstract: Storm surge is a primary driver of coastal flooding, induced by changes in mean sea level pressure and winds from tropical and extratropical cyclones. Sea level is observed both from land and from space: tide gauges provide high frequency records at fixed points, while satellite altimetry offers broad spatial coverage where in-situ instruments are sparse. Neither is complete on its own, as gauge records vary in length and coverage thins outside the Northern Hemisphere, while altimeter revisit times cause many surge peaks to be missed. Reconstructing surge from these observations conventionally requires hydrodynamic models, which are computationally expensive to run at the resolution required for coastal applications. We present a scalable deep-learning and high-performance computing framework for reconstructing storm surge along the global coastline by integrating tide-gauge observations, satellite altimetry, ocean and atmospheric reanalysis, and coastal characteristics. Rather than developing an independent model for each location, a global long short-term memory (LSTM) model is trained across tide gauges using static attributes such as shelf width, tidal range, nearshore slope, and coastal exposure to represent local characteristics. Models are trained on GPUs using ACCESS resources on Purdue’s Anvil supercomputer. The framework processes heterogeneous global coastal datasets, with tidal decomposition, and storm surge extraction from tide gauges parallelized using SLURM job arrays. Multi-mission along-track satellite altimetry spanning 1992-2024 is processed separately to derive nontidal residuals at coastal locations, providing an additional observational source in regions with limited gauge coverage. Atmospheric and ocean predictors from 1940-2024 are accessed directly from cloud-hosted ERA5 Zarr stores, reducing data transfers and preprocessing bottlenecks. This framework enables global surge reconstruction and extension of records from 1940’s to present into data-sparse coastal regions, and support flood hazard assessments, infrastructure planning, and long-term adaptation.
Dr. Abdelkader Kara, Artificial Intelligence based Modeling for Materials Discovery and Design (AIM4MD2)
Abstract: Computational materials’ studies relied heavily on density functional theory (DFT). Materials’ discovery was mainly based on trial and error or high-throughput calculations. Both were computationally expensive with no guaranteed success at the synthesis and characterization levels. Materials’ design, or inverse design, where a material is built atom by atom to satisfy a priory defined properties, was just a dream a few years ago. With advent of the large variety of AI tools, it is possible to combine seamlessly both design and discovery of novel materials with enhanced and well targeted properties. In this talk, I will present two examples where AI tools are used to discover and design novel materials. The first one consists of discovering new unseen high entropy alloys (HEA) clusters of different sizes, with over five elements and various stoichiometries. The newly discovered materials are then tested for use as hydrogen storage and for catalysis. The second approach consists of using AI models to designing novel materials targeted at enhancing the voltage and capacity of cathodes for lithium-ion batteries (LIB).
Now is the time with the paradigm shift from an experimentalist asking a computational scientist to determine the properties of a given material to a computational scientist telling an experimentalist ‘Tell me what you want, I have it’.
Shiyu Zhang representing Dr. Mohsen Rakhshan’s lab, Listening Motor Neurons Through the Skin: Decoding and Tracking Motor Neuron Activity with Research Computing
Abstract: Every voluntary movement you make is produced by motor neurons (MNs). These cells are located in the spinal cord, each one is connecting to a group of muscle fibers to form a motor unit, and they are the sole route through which the brain acts on the body also the final common pathway of the nervous system. Because every descending command finally converges on the MNs, the timing of their electrical discharges is a readout of neural control itself. Changes in how motor neurons fire underlie fatigue, aging, and recovery, and are altered in stroke, spinal cord injury, and neurodegenerative disease.
Studying them requires two capabilities. The first one is listening: identifying when individual motor neurons discharge. The second is tracking: following the same motor neurons over time. Identifying the MN discharge timing help us understand how our MN works, while tracking enable us to understand the changes in our PNS across hours, days, weeks, months or even years.
Traditionally, listening meant inserting needle or fine-wire electrodes into the muscle. This invasive method is uncomfortable, and only samples only a small number of motor units near the electrode. To overcome these difficulties, modern researchers utilize the non-invasive High-density surface electromyography to conduct their research, but each electrode captures the overlapping activity of many motor units simultaneously. The measurement becomes a mixture of many MNs, and heavy computation is required to decode their individual activities. Matching motor units across experiments on different days is likewise a computational problem. Generally speaking, computation is what enables us to study motor neurons non-invasively.
Faculty Invited Talks – Session II
15-minute presentations
Dr. Samik Bhattacharya, Simulating the Invisible: HPC-Driven Discovery of the Root Cause of Turbine Exhaust Vibration
Abstract: Industrial gas turbine exhaust diffusers can suffer unsteady vibration when strut wakes impinge on downstream airfoil-like manway blades, but isolating the mechanism experimentally is difficult. We combined a scaled experimental rig with high-performance computing to resolve this interaction. Unsteady RANS and detached-eddy simulations, run on HPC resources, captured the full three-dimensional, time-resolved vortex shedding from the struts that pressure taps alone could not reveal. These simulations reproduced the narrow-band pressure spectra measured experimentally, confirmed the leading-edge separation seen in oil-flow visualization, and let us track how shed vortical structures move downstream and impinge on the manway leading edge. This let us test upstream geometry modifications in simulation before committing to hardware, showing that redirecting the shear layers breaks down the coherent structures and eliminates the pressure spikes. HPC was essential not just for validating the experiment, but for diagnosing the root cause and screening a design fix before physical testing.
Dr. Veeraraghava Hasti, Artificial Intelligence-Driven Adaptive Digital Twins for Industrial Applications
Abstract: AI-driven adaptive digital twins represent a promising direction for improving the reliability, efficiency, and sustainability of industrial systems. This talk provides a high-level overview of the concept and the technologies needed for its future realization. It discusses how physics-based modeling, high-fidelity simulation, operational data, domain knowledge, and machine learning could be integrated to develop digital twins that learn and adapt as operating conditions change. Representative industrial use cases will be discussed.
Dr. Mengjie Li, From Knowledge Graphs to a Cloud-Hosted Benchmark: Ontology-Grounded Spatiotemporal AI for Power-Outage Resilience
Abstract: Resilient energy infrastructure depends on place-based decisions, yet the data needed to make them—outage records, satellite nighttime lights, storm reports, and social vulnerability indices—remain siloed and correlation-based. This talk traces our work from data integration to a deployed, cloudhosted research service. GeoOutageKG (ISWC 2025) is a multimodal geospatiotemporal knowledge graph that unifies over 10 million county-level outage records, 300,000 nighttime-light satellite images, and 15,000 high-resolution outage maps under a shared ontology, enabling multiresolution outage analysis. Building on it, GeoOutageBench (ACM SIGSPATIAL 2026) benchmarks ambiguityaware, ontology-grounded knowledge graph question answering: how faithfully large language models translate ambiguous natural-language questions into SPARQL, how much ontologies improve interpretation, and how accurate multimodal retrieval is. We use Google Cloud to host a public demonstration of this work at geo-resilience.com, making our knowledge graph and benchmark openly explorable and giving the research community useful tools for infrastructure resilience analysis.
Dr. Daniel Ram, From Transcriptomes to Mechanisms of Immune Dysfunction: Elucidating Alternative Splicing in Viral Infection, Cancer, and Aging via High-Performance Computing
Abstract: Alternative splicing transforms ~20,000 human genes into over 140,000 isoforms, generating transcriptomic diversity that enables mammalian adaptation. Dysregulation of splicing contributes to cancers, immune dysfunction, and altered host responses during infection. Using HPC, we analyze short- and long-read RNA-seq from human and macaque PBMCs and tissues (including hippocampi) across age, sex, and retroviral infection (HIV/SIV), and examine immune cell contact dependent modulation of splicing in tumor microenvironments. Our work entails full-length isoform detection, minor-intron annotation, differential splicing analysis, and network analysis, only possible with HPC resources. Findings include cell type-specific expression of minor intron-containing genes/isoforms; HIV/SIV-driven isoform remodeling; and sex- and age biased splicing networks in the aging hippocampus linked to inflammation and oxidative phosphorylation. Differential isoform expression also appears contact-dependent in the TME. These HPC-enabled analyses will allow us to establish an atlas of alternative splicing and its contribution to immune, neuroimmune, and tumor–immune dysfunction.
Dr. Patrick Meagher, A Hidden Fluidic Jet at the Heart of Detonation Cells
Abstract: Gaseous detonation waves, whether harnessed for hypersonic propulsion or occurring naturally in a Type Ia supernova explosion, are universally characterized by the repeating shock wave structures called detonation cells. Since their first observation in 1959, the physical mechanisms responsible for cell formation have remained elusive despite decades of advances in experimental diagnostics. In this work, high-fidelity numerical simulations enabled unprecedented examination of the origins of the detonation cells, revealing a high-speed (km/s) microscopic fluidic jet at the inception of each cell. While direct experimental observation of the fluidic jet remains impossible, high-performance computing has enabled us to quantitatively established its role in focusing energy into the surrounding unreacted gas, initiating nascent cell formation and sustaining the global detonation wave.
Faculty Invited Talks – Session III
15-minute presentations
Brook Miller, A Multi Model Approach to Historical Handwriting Transcription
Abstract: This talk reviews recent findings on the efficacy of different model types and prompting strategies in transcribing historical handwriting images to text. It notes disagreement about the utility of smaller models, then describes a developing project to allow researchers the ability to experiment with different models and prompting strategies with small archival datasets. This project, which is currently in the planning stages, is a partnership between the Office of Research Cyberinfrastructure and the Center for Humanities and Digital Research to build a test system that enables researchers to perform transcription with frontier and local models simultaneously and compare the results. Via the expertise in OR and the help of a CECS Senior Design programming team, the system will use an AWS implementation to manage batch processing and API calls to frontier models, as well as to coordinate calls to locally hosted and fine-tuned models on NSF’s Jetstream2 service.
Dr. Wayesh Qarony, Inverse Design Modeling of Spin–Photon Interfaces for Quantum Information Processing
Abstract: Quantum photonic circuits require efficient interfaces between solid-state qubits and photonic components while maintaining compact footprints, high performance, and fabrication compatibility. In this talk, I will discuss how computational inverse design can efficiently optimize the key building blocks of integrated spin–photon quantum systems. Using adjoint-based electromagnetic optimization, we design nanobeam cavities that strongly enhance light–matter interactions through Purcell enhancement while efficiently coupling photons from solid-state spin–photon qubits into guided optical modes. We further apply inverse-design approaches to compact photonic splitters and directional couplers for on-chip photon routing and quantum interference. I will present simulation/experimental results of nanophotonic devices integrated with spin–photon qubits on chip. These results demonstrate how high-performance electromagnetic modeling and inverse design can accelerate the development of fabrication-ready, high-performance quantum photonic components and their integration into scalable circuits for quantum information processing.
Dr. Chait Rendu, Unified Clinical‑AI Ecosystem on AWS (Training on HiPerGator, Inference on AWS)
Abstract: MammoVerse is a HIPAA-compliant, AI-powered clinical intelligence platform that combines patient engagement, multimodal reasoning, and personalized care navigation within a secure AWS-native ecosystem. The platform integrates patient intake, clinical data, imaging, knowledge graphs, and retrieval-augmented generation (RAG) to deliver contextualized guidance throughout the care journey. Models are trained on the HiPerGator high-performance computing environment and deployed through Amazon Bedrock and SageMaker for scalable inference and agentic orchestration. A personalized knowledge graph supports evidence-based recommendations, while a Hyperledger Fabric blockchain provides secure auditability and patient incentive management. This presentation describes the MammoVerse architecture and its role in enabling scalable, trustworthy, and patient-centered AI for healthcare.
Dr. Eduardo Mucciolo, A Quantum Algorithm for Determining Exchange-Correlation Potentials
Abstract: In this talk, we will present a quantum algorithm for determining the spin-resolved exchange-correlation (XC) potential in electronic many-body systems. The XC potential plays a key role in the accuracy of density functional theory, which is the most widely method for computing the electronic properties of materials. We simulated our proposed quantum algorithm in cloud computing environments and obtained results that are competitive against other, more established but heuristic methods. Our study indicates that the method scales favorable with system size, indicating that hybrid quantum-classical algorithms remain a promising approach to tackle many-body problems in the next generation of quantum computers.
Nurun Naher representing Dr. Gita Sukthankar’s lab, Scaling Qualitative Research: A Multi-Agent LLM Workflow for Thematic Analysis
Abstract: Qualitative analysis is the systematic analysis and interpretation of non-numerical data. One qualitative method commonly used by social scientists is thematic analysis, a technique for identifying and organizing patterns of meaning across a dataset. However, research studies can generate large collections of lengthy interview transcripts that require substantial time and effort for experimenters to organize, code, compare, and synthesize. Large language models (LLMs) offer opportunities to support and scale these analytic tasks. We developed a multi-agent LLM-assisted workflow for the thematic analysis of qualitative interviews collected by the IN STEP Children’s Mental Health Research Center. The workflow uses multiple computational stages to process transcript segments, generate and refine codes, organize themes, and connect themes with supporting evidence. This work explores how LLM-based methods can make qualitative analysis more scalable while examining the robustness, grounding, and reliability of generated interpretations.