WKU 5th Annual Data Science Day
We invite you to join us for the 5th annual WKU Data Science Day, one of the many great events occurring during WKU Libraries' Love Data Week celebration. Data Science Day is a full day of great presentations from current students, faculty, alumni, and campus partners like those at the Innovation Campus taking place on Thursday, February 12, 2026 at The Commons at Helm Library, 11:00am - 6:00pm.
11:20am - 11:30am
Dr. Cathleen Webb, Director of Applied Research and Technology Program and Dr. Richard
Schugart, Director of Applied Center for Data Science
Opening11:30am-12:30pm
Manil Maskey, Senior Research Scientist and Manager of the Office of Data Science
and Informatics, NASA’s Marshall Space Flight Center
First Planetary Talk, Room 3002/8
Data Science at NASA: Accelerating Discovery
Modern scientific discovery is increasingly data-driven, allowing for the optimization
of the scientific workflow and unlocking new discoveries through state-of-the-art
data science methodologies. This talk will describe how data science is empowering
NASA Science to enable accelerated discovery, from the development of domain-specific
AI foundation models and multimodal learning systems to open science platforms and
data-centric AI that enhances scientific workflows. The talk will also highlight the
opportunities and challenges of applying data science to the scientific domain, and
how we might work together to advance even further towards enabling accelerated discovery
for the research community.
Dr. Manil Maskey is a Senior Research Scientist and Manager of the Office of Data
Science and Informatics at NASA’s Marshall Space Flight Center. An expert in data
systems, AI, and advanced visualization, Dr. Maskey has over 25 years of experience
working in industry, academia, and government. Prior to joining NASA, he worked on
large-scale data mining applications at NCR/Teradata and machine learning research
for NASA and NSF at the University of Alabama in Huntsville (UAH).
At NASA, he has played a leading role in science data system modernization, cloud
initiatives, and has led high-profile competitive solicitations, cross-agency programs,
and international projects. He was previously the Data Science and Innovation Lead
for NASA’s Science Mission Directorate, where he led the agency’s efforts in AI. He
also established the data management project for NASA's Commercial Satellite Data
Program. He is a recipient of the NASA Agency Exceptional Achievement Medal, NASA
Agency Silver Achievement Medal, several NASA Agency Group Achievement Awards, and
multiple NASA Marshall Innovation awards.
In addition to his work at NASA, he is an adjunct faculty member in the UAH Atmospheric
Science Department. He holds leadership roles in several organizations, including
senior membership in the Institute of Electrical and Electronics Engineers (IEEE),
chairing the Geoscience and Remote Sensing Society (GRSS) technical committee. He
represents NASA in interagency activities in the Big Data Interagency Working Group
and serves on the Steering Committee of the National Artificial Intelligence Research
Resource (NAIRR) Program.
Room: Helm 3002/8
12:30pm-12:50pm
Shane Holinde, Outreach Manager, Kentucky Mesonet
Kentucky Mesonet: Measuring Weather, Establishing Climate
Information on how Kentucky Mesonet, including the history of the network, its mission,
and how the organization uses various sensors and instrumentation at its 86 stations
to measure various weather parameters in real-time across the state. Data collection
beneficial to agriculture and horticulture interests, including measurements from
soil probes along with near-surface temperature inversions, will be explored. Utilization
of weather data to comprise an official climate record for the Commonwealth along
with its usefulness in research for applied climatology and seasonal trends will also
be explained.
Room: Helm 3002/3008
12:50pm-1:00pm
Henry Lee, Student Researcher, The Gatton Academy of Mathematics and Science
Evaluating Renewable Energy Growth Dynamics and Their Implications for Climate Change
Mitigation
The rapid expansion of renewable energy is central to climate change mitigation strategies.
However, many long-term projections rely on sustained exponential growth, potentially
overlooking real-world structural, economic, and infrastructural constraints. This
project proposes a data-driven evaluation of whether historically observed renewable
energy adoption trends align with the scale and pace required for effective climate
mitigation. Using publicly available data from Our World in Data on renewable energy’s
share of total energy consumption, this study applies mathematical growth modeling
to country-level and global time series. Exponential and logistic growth models are
fitted using nonlinear least squares methods, and their performance is compared using
root mean square error (RMSE) and standard model fit metrics. Particular attention
is given to identifying saturation effects and estimating implied upper bounds on
renewable energy penetration. The analysis aims to assess how different growth assumptions
influence projections of future adoption and to compare modeled trajectories with
commonly cited energy transition benchmarks. By highlighting discrepancies between
historical trends and climate mitigation targets, this work demonstrates how statistical
modeling can inform more realistic assessments of large-scale energy transitions.
The proposed presentation will emphasize methodology, model interpretation, and implications
for climate policy and planning, showcasing how data science tools can be applied
to complex, real-world sustainability challenges.
Room: Helm 3002/8
1:00pm-1:15pm
Break
1:15pm-1:25pm
Dr. Belinda J. Petri, Post-doctoral Fellow, Kentucky IDeA Networks of Biomedical Research
Excellence Bioinformatics Core, Department of Biochemistry and Molecular Genetics,
University of Louisville School of Medicine
1:25pm-1:35pm
Vedant Garg, Student Researcher, The Gatton Academy of Mathematics and Science
Prediction of glioma tissue stiffness using metabolomic signatures
Gliomas are aggressive tumors in critical need of improved therapeutic options. Recent
work has demonstrated that glial tissue from core (inner) and edge (infiltrating)
regions possesses distinct metabolic signatures and biomechanical properties that
are linked to tumor aggression and migration. In this proof-of-concept study, Young’s
moduli (stiffness) of core and edge tissue are predicted using paired metabolic signal
intensities.
Core and edge stiffness previously measured from n = 25 patients were paired with
metabolomic data previously obtained using 2D liquid chromatography-mass spectrometry/mass
spectrometry. Low (≤median) and high (>median) stiffness were predicted from paired
core and edge metabolomics using a machine learning (ML) workflow that included forward
feature selection, model training, grid search hyperparameter tuning, and repeated
k-fold cross-validation.
Key core metabolites predictive of low and high stiffness in core tissue included
N6-methyllysine, 2',3'-cyclic UMP, and gamma-amino-n-butyric acid. Top core metabolites
in predicting edge moduli included guanosine, acetylcholine, glutamic acid, and N6-methyllysine.
The top edge metabolite in predicting edge moduli was DL-p-hydroxyphenyllactic acid.
Using ≤5 features, machine learning models predicted core and edge moduli using core
and edge metabolites individually and in combination, achieving AUROC, maximum F1,
and PRAUC values ≥0.90.
This study shows that regions of differing glioma core and edge stiffnesses exhibit
unique metabolic signatures. These signatures could potentially be explored to develop
personalized therapeutic strategies.
Room: Helm 3002/8
1:35pm-1:55pm
Aaron Owens, Skylight Digital
Emerging AI Threats to Data Integrity
AI is increasingly being used to steal and manipulate data as well as spread false
information. Former US Air Force Cyber Warfare Officer and current Cybersecurity Engineer,
Aaron Owens, will give an overview of these emerging threats to data integrity. This
look into the future of Data Protection and Cybersecurity will be invaluable in helping
you plan your career.
Room: Helm 3002/8
1:55pm-2:40pm
Poster Session
M Zakaria Kurdi, Meharry Medical College Faculty *CANCELED*
LaPorchia Davis, Graduate Student Research, Meharry Medical College
Integrating Clinical Predictive Modeling and Image-Based Analysis to Identify Bone Health Complications in Burn Patients
Integrating Clinical Predictive Modeling and Image-Based Analysis to Identify Bone Health Complications in Burn Patients
Severe burn injuries often trigger musculoskeletal complications that extend far beyond
skin damage. Many survivors experience “hidden” bone-related conditions such as fractures,
bone mineral density loss, and osteomyelitis, which may not be easily detected through
routine clinical assessment alone. Because these complications can progress silently
and lead to long-term skeletal deterioration, early prediction and targeted monitoring
are essential for preserving mobility and quality of life.
This project presents a multimodal approach that integrates clinical predictive modeling
with advanced imaging analytics to evaluate bone health complications in burn patients.
The clinical component analyzes an AI-generated synthetic dataset containing 1,538
burn patient records using predictive modeling and principal component analysis (PCA)
to identify key factors linked to fractures and osteomyelitis. The imaging component
uses deep learning–based feature extraction (ResNet-50) combined with Uniform Manifold
Approximation and Projection (UMAP) to explore structural variation across a 72-slice
Computed Tomography (CT) from 5K+ CT Images on Fractured Limbs dataset.
Our preliminary findings show that burn severity index, bone damage score, duration
of hospital stay, comorbidity score, and treatment material properties are the strongest
predictors of fracture risk and osteomyelitis development. Additionally, the CT slice
embeddings form distinct anatomical clusters, indicating that the model captures meaningful
structural cues that may support automated bone health assessment.
Overall, this multimodal framework combining clinical features with CT scans through
deep-learning embeddings is hoped to advance early detection strategies for burn-related
skeletal complications and help address persistent gaps in integrating clinical and
imaging data for bone health prediction.
Kendall Al-Halah, Student Researcher, Department of Psychology, Student in Data Science
and AI Travel Grant Recipient, Western Kentucky University
Blind Loyalty to the Sunk-Cost Decision of a Loyal Leader
Blind Loyalty to the Sunk-Cost Decision of a Loyal Leader
Leadership success depends not only on leader characteristics but also on how group
members respond to leader behavior. One important member trait is group loyalty, defined
as emotional attachment and commitment to one’s group. Loyal members are typically
expected to support collective goals and help leaders make sound decisions. However,
loyalty may have unintended consequences when leaders make irrational choices, such
as those driven by the sunk cost fallacy, where individuals continue investing in
failing projects to justify prior investments. Prior research offers mixed conclusions,
suggesting that loyalty can either blind members to poor leadership or motivate dissent
to protect group welfare. We propose that these outcomes depend on how members perceive
the leader’s own group loyalty. Drawing on trait activation theory and the similarity-attraction
principle, we argue that leaders who display group loyalty through commitment and
self-sacrifice activate members’ dispositional loyalty, increasing trust and conformity
to the leader's irrational decision. Across two online experiments (N = 330), undergraduate
participants evaluated a university leader facing a sunk cost decision. Results showed
that member group loyalty predicted conformity to the leader’s sunk cost decision
only when the leader was described as loyal; this relationship disappeared in a control
condition without leader loyalty cues. These findings suggest that a combination of
loyal leaders and loyal members—often viewed as ideal—may increase vulnerability to
biased decision-making, highlighting a potential downside of loyalty in group and
organizational contexts.
Room: Helm 3002/8
2:40pm-3:00pm
Benjamin Blankartz, AMEND Consulting, Logan AluminumAdam Seal, AMEND ConsultingJeff
Beal, Logan Aluminum
Machine downtime and hidden bottlenecks are expensive—and in high-throughput manufacturing,
minutes matter. In this session, Logan Aluminum and AMEND Consulting share how they
built and deployed a production IoT Digital Twin Simulation and Predictive Planning
Tool that is delivering millions in higher throughput by turning plant-floor IoT signals
into forward-looking decisions. Rather than reacting when a line backs up or a critical
asset goes down, the Digital Twin tracks product movement across the plant in near
real time, predicts process times, and uses discrete event-driven simulation and what-if
scenario analysis to forecast where flow will constrain—days before it happens. Planners
can test interventions (reroutes, staffing changes, schedule adjustments, work-center
speed changes) and choose the option that best protects throughput and reduces downtime
risk, all delivered through Microsoft’s Power Platform for fast adoption. We’ll also
show how the same model scales beyond daily operations into strategic capacity planning:
evaluating capacity improvement, predicting bottleneck shifts, and quantifying ROI
before capital is committed. Attendees will leave with a practical blueprint for building
industrial grade digital twins—data foundations, modeling approach, deployment considerations,
and the lessons learned that made this one stick.
Room: Helm 3001
2:40pm-3:00pm
Khalil Garmin, Founder of Moneybot
Room: Helm 3008
3:00pm-3:10pm
Madison Gardner, Student Researcher, Undergraduate Student in Data Science Grant Recipient,
Student in Data Science and AI Travel Grant Recipient, School of Engineering and Applied
Sciences
A Real-Time, Learning-Enabled IIoT Framework for Multi-Sensor Robotic Perception in
Smart Manufacturing
Smart manufacturing environments increasingly rely on heterogeneous sensor networks
(robot controllers, industrial cameras, and embedded IMUs) to enable adaptive, data-driven
automation. However, integrating these data streams into a unified analytics pipeline
remains a significant challenge for small and mid-scale laboratories. In this work,
we present a real-time Industrial Internet of Things (IIoT) framework that fuses robot
telemetry, camera-based pose estimation, and IMU sensor data to support AI-driven
predictive maintenance and digital-twin development. Our system uses MQTT-based messaging,
containerized microservices, and Python-based data transformation to achieve reliable
latency. The resulting dataset is working to enable downstream machine-learning tasks
such as anomaly detection, tool-path deviation prediction, and multisensor state estimation.
Our architecture is working to support stable, high-frequency sensing while remaining
accessible to undergraduate research teams. This work highlights the role of data
science and AI in building scalable, transparent, and reproducible smart-factory testbeds.
Room: Helm 3001
3:00pm-3:20pm
JC Watkins, Kentucky Science & Technology Corporation
Where Federal, State, and Local Governments Use Data Scientists
Data science has become a foundational capability for modern government operations
at the federal, state, and local levels. As governmental agencies and consulting firms
manage growing volumes of data related to public safety, infrastructure, health, finance,
and service delivery, the need for advanced analytical skills continues to expand.
This presentation provides a practical overview of where and how governments use data
scientists to improve decision-making, optimize resources, and enhance mission outcomes.
Room: Helm 3008
3:10pm-3:20pm
Jacob Kennedy, Student Researcher, Undergraduate Student in Data Science Grant Recipient,
Student in Data Science and AI Travel Grant Recipient, School of Engineering and Applied
Sciences
IIoT and Real Time Predictive Maintenance Architecture
The Fourth Industrial Revolution (Industry 4.0) has introduced data-driven technologies
such as artificial intelligence, machine learning, and the Industrial Internet of
Things (IIoT) into modern manufacturing, enabling more advanced communication between
machines and processes. IIoT allows manufacturers to build interconnected networks
of machines, sensors, controllers, and devices that continuously transmit data, forming
the foundation for machine learning applications and predictive maintenance systems.
Predictive maintenance leverages historical process data to identify patterns that
precede equipment failures, allowing manufacturers to anticipate and prevent unexpected
downtime. My presentation will detail how a campus-based IIoT was implemented by our
research teams, which includes robots, controllers, cameras, computers, remote I/O
modules, and a dedicated server rack. In collaboration with undergraduate and graduate
teams. The researchers enhanced a FANUC robot with inertial measurement units and
ArUco markers to create a digital twin in Unity. Sensor data from the physical robot
was compared with the digital twin to verify positional accuracy and track deviations
during movement. By analyzing discrepancies between expected and actual positions
over time, engineers can predict performance variance and identify when a robot may
fall outside manufacturer specifications. This approach demonstrates the core principles
of predictive maintenance and highlights how such systems can be developed effectively
within an academic setting using cost-efficient hardware such as Raspberry Pi–controlled
sensors connected to a local network.
Room: Helm 3001
3:20pm-3:40pm
Jeremy Maddox, WKU Faculty, Department of Chemistry
Visualizing Quantum Dynamics with GWPTools: A Mathematica Framework for Gaussian Wavepackets
At the microscopic scale, matter behaves according to the laws of quantum mechanics.
The state of a system—such as a single particle or a complex molecule—is represented
by a wavefunction that evolves over time according to the Schrödinger equation. In
computational physics and chemistry, Gaussian Wavepackets (GWPs) are frequently used
as a foundational tool to simulate these dynamics because they maintain a predictable
mathematical structure under specific conditions. This presentation introduces GWPTools,
a Mathematica package designed to streamline the manipulation and visualization of
generalized, complex-valued GWPs. We will demonstrate how this package automates the
calculation of time-dependent parameters and enables the generation of various kinds
of plots to illustrate quantum behavior. By bridging the gap between abstract equations
and visual data, GWPTools provides an accessible framework for researchers and students
across disciplines to explore quantum systems.
Room: Helm 3001
3:20pm-3:40pm
JD Thomason, Data Advocate, Houchens Industries
Why Do Most Data Projects Fail?
Classes teach you to work with clean datasets and clear requirements. The real world
is messier. At both Walmart and Houchens Industries, I’ve learned that success depends
less on technical perfection and more on navigating ambiguity - conflicting data sources,
unclear business requirements. Drawing from projects ranging from AI-powered product
onboarding at Fortune 1 scale to data governance challenges at regional operations,
I’ll share practical lessons about what data work actually requires and how students
can develop the skills that matter most in industry.
Room: Helm 3008
3:40pm-3:55pm
Break
3:55pm-4:05pm
Michael Seavers, Graduate Student Researcher, Student in Data Science and AI Travel
Grant Recipient, School of Engineering and Applied Sciences
Identifying the Optimal Number of DataLoader Workers for CPU-GPU Concurrency in Asynchronous
Deep Training Pipelines
Efficient CPU-GPU concurrency is critical for maximizing performance in deep learning
training pipelines, yet selecting the optimal number of DataLoader workers remains
largely empirical. This paper introduces a framework for identifying the optimal worker
count in asynchronous deep training pipelines by integrating a new architectural metric,
SM:CC:W—the ratio of GPU Streaming Multiprocessors (SMs), CPU cores (CC), and DataLoader
workers (W). By jointly analyzing elapsed time, GPU/CPU utilization patterns, and
workload distribution across major components of the data pipeline, the framework
provides a principled mechanism for understanding and optimizing CPU-GPU overlap.
We evaluate the approach on five single-machine systems with diverse CPU-GPU configurations,
including three local machines and two Google Colab platforms. Our results demonstrate
that the new metric, combined with the proposed performance analyses, reliably identifies
worker counts that maximize GPU utilization while avoiding CPU bottlenecks. Beyond
guiding DataLoader tuning, the framework offers broader insights into how architectural
characteristics shape concurrency behavior in modern deep training systems.
Room: Helm 3001
3:55pm-4:05pm
Alexander Bentley, Student Researcher, Department of Analytical & Information Systems
Program-Level Data Collection of Student Outcomes in the Kelly Autism Program
The Kelly Autism Program (KAP) at Western Kentucky University is a comprehensive support
program design to promote academic success, independence, and well-being for autistic
college students. From Spring 2025, KAP has generated a large body of program-level
data through routine advising, mentoring, academic supports, wellness check-ins, and
structured programming. This project brings that data together to explore and reveal
how autistic students engage with, benefit, and use support in higher education.
Using de-identified internal program data, this analysis examines patterns related
to student engagement, service utilization, and retention over time. Data sources
include attendance records, advising and mentoring logs, participation, and program
outcome measures. Rather than focusing on deficit based metrics, this work measures
quantitatively on strengths, consistency of needed support, and the role of structured
interventions in student success.
This information demonstrates how data collected as part of an everyday program operations
can be transformed into meaningful evidence to inform practice, improve services,
and support advocacy for higher education models. Findings offered will provide practical
insights for autism support programs, student service professionals, and institutions
seeking data-driven approaches to accessibility, inclusiveness, and student persistence.
Broader speaking, this project works to uncover valuable information already embedded
within existing support programs and highlights shifts to evaluate success for autistic
college students.
Room: Helm 3008
4:05pm-4:25pm
Peter Agaba, Senior Analyst, Mass General Brigham, Western Kentucky University alumnus
Leveraging Machine Learning for Early Detection and Management of Hemoglobin A1c in
Diabetic Patients
There are more than 38.4 million people living with diabetes in the USA, which translates
to 11.6% of the U.S. population. Diabetes is the eighth leading cause of death in
the U.S. Effective management of Hemoglobin A1c (HbA1c) levels is crucial in reducing
diabetes-related complications and improving patient outcomes. This paper explores
the application of advanced machine learning (ML) and statistical techniques for the
early detection and management of HbA1c levels in diabetic patients. Specifically,
it classifies HbA1c levels of less than 8% as good control and levels greater than
9% as poor control. By classifying patients as high-risk versus non-risk using analytics,
the paper recommends personalized treatment plans. This approach enhances glycemic
control and advances precision medicine. The proposed system aims to offer diagnostic
and treatment recommendations tailored to individual patients while implementing remote
monitoring to reduce hospital readmissions and optimize healthcare resources. This
works aims to transform diabetes care, improve patient outcomes, and reduce healthcare
costs.
Room: Helm 3001
4:05pm-4:15pm
Ama Boateng, Graduate Student Researcher, Student in Data Science and AI Travel Grant
Recipient, Department of Psychological Sciences
Paternal Intrusiveness: The Roles of Parent Efficacy, Temperament, and Support
Parental intrusiveness, or the degree to which parents impose their own agenda on
their child, can hinder children's emotional and behavioral outcomes (Eisenberg et
al., 2015; Smaling et al., 2017). Parent efficacy and infant temperament have been
more commonly examined with other parenting behaviors (e.g., parental sensitivity),
but the extent to which they predict parent intrusiveness is unclear (Bailes et al.,
2024; Grimes, 2012). Other predictors of parental intrusiveness, such as perceived
partner support, is largely unexamined (Razurel et al., 2016). Past research examining
paternal parenting behaviors is also limited (Volling & Cabrera, 2019). The present
study (n=62 dyads) examined the association between parental efficacy and paternal
intrusiveness across early infancy, including infant temperament and perceived partner
support as moderators. Fathers completed questionnaires assessing infant temperament
(Gartstein & Rothbart) and partner support (Cronenwett et al., 1988) when infants
were 4-months. Paternal intrusiveness was rated during a face-to-face parent-infant
play task when infants were 8-months (Braungart-Rieker et al., 2014; Tronick et al.,
1978). Multiple regression model results indicated direct care support (B=-.552, p˂.001)
and infant negative temperamental reactivity (B=.393, p=.012) were moderators between
parental efficacy and paternal intrusiveness. Although findings were consistent with
previous studies reporting that highly supported parents are more efficacious (Leahy-Warren
et al., 2011), our findings indicated they are also more intrusive. Perhaps highly
supported partners are less involved with their infants and subsequently less intrusive
(DePasquale & Gunnar, 2020). Future studies should continue to examine infant and
environmental factors that can affect fathers’ parenting behaviors.
Room: Helm 3008
4:15pm-4:25pm
Sujan Adhikari, Graduate Student Researcher, Student in Data Science and AI Travel
Grant Recipient, Department of Earth, Environmental, and Atmospheric Sciences
AI-Driven Fault and Lineament Mapping from High-Resolution LiDAR in Western Kentucky:
Random Forest and U-Net Workflows for Geologic Mapping and Seismic-Hazard Screening
Western Kentucky lies between the New Madrid and Wabash Valley seismic zones, yet
remains poorly characterized for seismic hazard. We use 1.5-meter resolution LiDAR
digital elevation models (DEMs) to build an AI-enabled, repeatable geospatial workflow
for fault/lineament mapping that also leverages hydrologic and geomorphic signals
of structure. We construct a multi-scale geomorphometric stack from the DEM (multi-azimuth
hillshade composites, slope, curvature, roughness, and topographic position indices)
and test three automated fault-detection approaches. These include (1) an interpretable
Random Forest classifier trained on geomorphic indices, (2) a U-Net convolutional
neural network for multi-channel semantic segmentation of lineament patterns, and
(3) a hybrid pipeline that integrates classic edge-detection filters with deep-learning-derived
feature maps. Model outputs are evaluated against existing state geologic maps, published
lineament interpretations, drainage-network attributes, and historical seismicity
to assess both geologic plausibility and geospatial performance. Across test areas,
the models improve the continuity and spatial confidence of candidate fault-related
lineaments and identify corridors where structural control on drainage is strongest.
We translate these results into uncertainty-aware mapping products that can be iteratively
refined with targeted field checks and new training labels. Detected structures are
incorporated into a preliminary seismic hazard screening layer that combines fault
likelihood with earthquake catalogs, site conditions, and exposure layers (infrastructure
and building footprints) to highlight zones of elevated vulnerability. Although motivated
by seismic-hazard reconnaissance in a data-limited intraplate setting, the workflow
and derived products are directly transferable to hydrologic and geomorphic mapping,
including delineation of structurally influenced drainage corridors, valley-lineament
networks, and terrain units from LiDAR-based surface metrics.
Room: Helm 3008
4:25pm-4:45pm
Lukun Zheng, WKU Faculty, Department of Mathematics
The Future of Data Science in the Age of Generative AI
Generative AI is rapidly transforming data science by reshaping how we collect, analyze,
and communicate data. This presentation offers a clear overview of how large language
models and AI-driven tools streamline workflows such as data cleaning, exploratory
analysis, modeling, visualization, and reporting. It highlights the new skills students
need in a GenAI-enhanced environment—from problem framing and critical thinking to
evaluating AI-generated code, insights, and explanations. Through practical examples
and brief demonstrations, the talk will show both the capabilities and limitations
of generative AI, including issues of accuracy, bias, and responsible use. Attendees
will gain an accessible, forward-looking understanding of how generative AI is changing
expectations for data scientists and how students, educators, and researchers can
strategically integrate these tools to enhance learning, productivity, and discovery.
Room: Helm 3001
4:25pm-4:45pm
Preston Burns, Sarepta Therapeutics
Visual Analytics for Drug Safety: Modernizing Safety Review with R/Shiny
Safety review in drug development is often slowed by static tables and long listings,
making it difficult to identify urgent safety signals across adverse events, labs,
dosing, and concomitant medications. This talk highlights how modern data science
approaches - particularly R/Shiny applications, reproducible pipelines, and human‑in‑the‑loop
visual analytics - are transforming safety monitoring in the pharmaceutical industry.
We will explore interactive safety graphics and visual patient profiles that allow
reviewers to fluidly shift between aggregate trends and patient‑level detail. Model‑assisted
workflows such as alerting, outlier exploration, and cohorting will be discussed alongside
the practical data engineering that supports them, including standardized inputs,
predictable refresh cadences, and lightweight metadata for cross‑study scalability.
The talk will also cover organizational considerations such as navigating exploratory
vs. GxP contexts and will close with real‑world lessons learned and key challenges
encountered during implementation.
Room: Helm 3008
4:45pm-5:00pm
Break
5:00pm-6:00pm
David Ciommo, Data Storyteller, Humana
Data Storytelling: The Data Literacy Skill Bridging the Human & Machine Intelligence
Gap
In an era where AI reshapes how we interpret information, data storytelling has become
an essential data literacy skill for making faster, more confident decisions. This
session explores how transforming complex data into compelling, human‑centered narratives
elevates Decision Intelligence and drives sustainable business growth. David will
uncover strategies and best practices for leveraging AI‑powered insights in ways that
are accessible and impactful across every level of an organization. We’ll examine
how the brain processes information and how purposeful storytelling naturally amplifies
our decision‑making abilities. Join David on a journey where human intuition and computational
intelligence converge—empowering you to navigate the digital landscape with clarity,
confidence, and a renewed sense of possibility.
David Ciommo is the Data Storytelling Lead and leader of the Visualization Center
of Excellence at Humana. In this role, he guides associates and executive leadership
in finding actionable opportunities through the data's story. His work has led him
to be recognized and recently awarded Data Storyteller of the Year by Narrative Science.
In addition to his daily responsibilities, David also speaks, educates, consults,
and mentors on topics including data literacy, data visualization, data storytelling,
decision intelligence, and design thinking.
Prior to Humana, David worked in the marketing, advertising, and communications industry
for over a decade and a half. He has contributed to and created some of the world's
most recognizable brands, such as The Discovery Channel’s original Animal Planet logo.
As Senior Art Director at the World Wildlife Fund, he led and designed countless products
and award-winning advertising campaigns. David’s expertise in visual storytelling
has contributed to many global campaigns for Lockheed Martin, Papa John's, and the
Milk Processor Education Program “GOT MILK?” campaign.
Room: Helm 3002/8
Some of the links on this page may require additional software to view.