100 Years of Innovation: Duke Graduate Research Fair Presentations

Image

Ph.D. student, Computational Media, Arts & Cultures

Advisor: Victoria Szabo

When the Body Becomes an Interface: A Wearable Artwork About Data, Touch, and Control

What happens when the technologies closest to our bodies turn bodily experience into data? 

My research examines how wearable devices, medical sensors, robotic systems, and intelligent environments reshape what it means to perceive, act, and have agency. These technologies can help us monitor health, communicate, and extend our senses. Yet they also translate touch, movement, and physiological processes into signals, measurements, predictions, and forms of control. I ask how we can make those hidden translations visible—and question who or what gains control when our bodies become interfaces. 

My project, Intimate Circuits, brings together my experience as a participant in biomedical research, archival study of historical medical instruments at Duke, and the construction of experimental artworks. I trace a longer history in which stethoscopes, electrotherapy belts, and pulse-recording devices converted bodies into sound, current, traces, and numbers—long before today’s smartwatches and AI systems. At the fair, I will wear an interactive wearable garment made from copper touch surfaces, exposed wires, sensors, microcontrollers, moving rods, and LED displays. When I touch it—and, where appropriate, when visitors touch designated surfaces—the wearable transforms contact into changing light patterns and movement. The interaction makes its process legible: body → sensor → code → feedback → body. Rather than treating technology as invisible or neutral, this live display invites visitors to see and discuss the intimate systems already surrounding us. It asks: What is gained, what is lost, and how might we design more caring relationships among bodies, machines, and environments?

Image

M.A. student, Graduate Liberal Studies

Advisor: Ralph A. Litzinger

From Shared Desks to Separate Futures: Life Histories of Migrant Peers

We all remember the classmates we sat next to in school, sharing pencils, jokes, and daily routines. While we have the privilege of pursuing higher education at Duke, what shaped the journeys of the peers who started at those same desks? Why did some enter the workforce early, marry young, or return to rural hometowns while others continued on to university? 

Drawing on 13 life-history interviews in China, my research examines how urban institutional barriers guide migrant children into divergent life paths. The findings show that policy thresholds impact entire households: when youth are restricted from local urban high schools, families often split, leading mothers to leave city jobs and return home as "accompanying mothers" to support their children's education. The presentation uses trajectory maps, and paired life timelines to invite visitors to reflect on their own school memories. By connecting individual life stories with structural realities, the project highlights the research value of life history.

Image

M.A. student, Graduate Liberal Studies

Advisor: Kent W. Wicker

Nation(alism) and Queer: Considering the Intersectional Experience of Chinese Queer Diaspora in London

Queer diaspora has recently received much academic attention, but its relationships with nationalism remain a topic warranting scholarly discussion. Through conducting 22 semi-structured interviews with Chinese queer diaspora in London, UK, this study examines their transnational experiences under the intersectionality of their queer identity and national identity, to consider two research questions: how could we understand the role of national identity and nationalism in queer diaspora’s transnational migration, and would it be possible for queer diaspora to become nationalists. 

This study finds that homonationalism can serve as a social catalyst to motivate queer people with available migrant resources to undertake migrant actions. After arriving in London, the queer diaspora who experienced othering under racialization and nationalization may challenge the Western homonormative queer identity and increase a national identification. In response to new social challenges, reactionary nationalism has emerged, which fosters new nationalist subjectivity and encourages social participation into transnational queer activism. Identity change is rarely one single identity change independently, but multiple identities intertwined with each other and change simultaneously. This study highlights the role of national identity and nationalism in queer study and migration study, and considering the nationalist side of current transnational queer activism.

Image

M.A. student, Religious Studies

Advisor: Daniel Herskowitz

Race-Making Among Jewish College Students: Duke University as Test Case

While definitions of race, ethnicity and origin shift and change over time, they continue to shape the lived experiences of everybody in the United States. The goal of my research is to better understand contemporary American Jews' relationship to race as it relates to their perception of what it means to be Jewish. To achieve this, I surveyed and interviewed the Jewish student population at Duke University about how racial hierarchy affected their sense of belonging both as Jews as well as racialized individuals within a diverse nation.

Image

Ph.D. student, Population Health Sciences

Advisor: Connor Drake  |  Collaborators: Stephen Greene

Food Is Medicine in Practice: Engagement, Dose-Response, and Differential Effects in Heart Failure

Heart failure is one of the leading causes of hospitalization in the United States, and many patients struggle to access the healthy foods needed to manage their condition. The Eat Well Heart Failure Trial is testing whether providing healthy food support can help prevent hospitalizations and deaths among patients with heart failure. This is the largest Food is Medicine trial conducted in patients with heart failure, enrolling 860 participants in just six months. 

The study uses a novel pragmatic trial design that brings research into real-world clinical care rather than requiring patients to seek out a traditional research study. We developed an affirmative outreach approach that proactively identifies and invites eligible patients to participate, helping us reach people who are often missed by conventional recruitment strategies. Community engagement is central to the study. Patients and community partners helped shape how we communicate about the trial, recruit participants, and deliver the intervention. By combining community-informed research with innovative trial methods, Eat Well HF is demonstrating how Duke can conduct clinical research that is faster, more inclusive, and more closely connected to the needs of patients. Our goal is not only to understand whether Food is Medicine improves heart failure outcomes, but also to create a model for how large, patient-centered trials can be conducted more effectively across health systems.

Image

Ph.D. student, University Program in Ecology

Advisor: Tong Qiu  |  Collaborators: Tate Commission, Apollo Fung, Zain Radwan, Yuqian Wang, Michael Wang, Zhuohong Li, Tong Qiu

High-Resolution Mapping and Attribution of Trees Outside Forests for Ecological Conservation in Dryland Landscapes

Trees outside forests (TOF), including scattered trees and small woody patches across savannas and drylands, provide critical ecological functions such as erosion control, microclimate regulation, fuelwood, and habitat connectivity. However, TOF remain poorly represented in existing satellite products and restoration monitoring systems. This gap is particularly evident in Malawi, where widespread land degradation and forest loss have prompted national commitments to restore trees over millions of hectares. Effective restoration, however, requires spatially explicit evidence of where TOF persist, how they are changing, and which environmental conditions favor their survival. 

To address this knowledge gap, we aim to integrate high-resolution satellite imagery and deep learning to map TOF and their changes over time, to identify their environmental and anthropogenic drivers, and to inform ecological conservation decisions in dryland landscapes. We developed an integrated framework combining 3-m spatial resolution PlanetScope imagery, approximately 80,000 manually annotated tree crowns, canopy-height data, and deep-learning-based crown delineation to quantify TOF from 2018 to 2025. We then used a spatial mixed-effects model to relate pixel-level trends in TOF cover to baseline tree cover, climate, topography, soil properties, population pressure and restoration activity. Finally, we integrated estimated driver effects to identify restoration opportunities and inform conservation strategies. Explicitly incorporating TOF into national monitoring could help distinguish areas requiring active recovery from those where existing woody cover warrants protection, improve the targeting of restoration investments, and strengthen the evaluation of restoration outcomes across dryland landscapes.

Image

Ph.D. student, Cell and Molecular Biology

Advisor: Akankshi Munjal

The Shapes Inside Us

What shapes us? Our bodies are composed of trillions of cells, each part of organs with vastly different shapes. Remarkably, this great diversity develops from a single cell, the fertilized egg, which multiplies and generates diverse daughter cells that somehow know how to make the shapes inside us. 

While we know a lot about the genes that define cell lineages in each organ, much less is known about how three-dimensional shape is born from these lineages. Shape arises from the coordinated efforts of many cells as tissues, which undergo dramatic shape changes as organs mature. A striking example of this is tissue fusion, wherein two tissues come close together, touch, and merge into one, a risky physical transformation that is required to shape intricate organs like the eye, heart, and spine. Understanding how this physical process interacts with genes to ensure proper shape is critical for addressing fusion defects, which underlie many widespread congenital diseases. 

In my research, I study fusion during the formation of semicircular canals in the inner ear, three interconnected tubes that allow us to sense balance. Canals form when pairs of tissue buds touch and fuse into pillars, partitioning the future canals. I discovered that bud touch is not only part of the fusion process but also induces gene activity that defines pillar shape; when touch-recognition is disrupted in zebrafish, fusion occurs but canals are malformed, and adult fish exhibit imbalance. This highlights that transient physical interactions, often overlooked, are important for organ form and function.

Image

Ph.D. student, Molecular Cancer Biology

Advisor: Ann Marie Pendergast

Uncovering The Contribution Of Stress Granules To TNBC Tumor Biology And Their Regulation By The ABL Kinases

Cancer cells acquire cellular phenotypes that promote growth and metastasis by way of adapting to physiological and pharmacological stressors in their microenvironment. Among the adaptive responses exploited by cancer cells is the formation of stress granules. Stress granules (SGs) are transient, cytoplasmic condensates of RNA and protein that form as the cell’s response to intrinsic (i.e., oxidative stress) and extrinsic stressors (i.e., hypoxia, radiation, chemotherapy). SG formation has been associated with cell survival, therapy resistance, and tumor growth. Interestingly, proteins that drive SG formation, G3BP1 and G3BP2, have been shown to regulate cell migration, invasion, and tumor initiation in triple-negative breast cancer (TNBC). However, it is unclear whether G3BP1’s and G3BP2’s SG functions contribute to these phenotypes. Although G3BP1/2-targeting drugs have been developed and validated in vitro, these drugs have not been approved for human use. The Abelson family of non-receptor tyrosine kinases (ABL1 and ABL2) are candidate regulators of SG formation as they are activated upon stress stimuli and contain intrinsically disordered regions, two common characteristics of SG-regulating proteins. 

Using FDA-approved, allosteric inhibitors against the ABL kinases, I have shown that ABL kinase inhibition decreases oxidative stress-induced and chemotherapy-induced SG accumulation in TNBC cells. Further, we found that ABL kinases regulate SG formation and localize SGs. I have also shown that high expression of SG proteins (G3BP2 and TIA-1) and the ABL kinases in brain-metastatic TNBC cells. Therefore, I hypothesize that ABL-mediated regulation of SGs can be a promising therapeutic strategy to prevent aggressive phenotypes in metastatic TNBC.

Image

Ph.D. student, Biology

Advisor: Susan Alberts

Vertical Vantage: Paired Cameras Reveal Cross Strata Predator-Prey Dynamics in Malagasy Rainforest

A short presentation on a canopy and terrestrial camera trap survey led with the goal of addressing a critical blind spot in primate monitoring: occupancy and dynamics of arboreal communities. By deploying a paired canopy and terrestrial camera trap matrix over 356 days in the Anjanaharibe-Sud Special Reserve, of northeastern Madagascar, we documented all known lemur species in the reserve, including the Critically Endangered Propithecus candidus and Indri indri. Crucially, our findings demonstrate that 62% of unique detections occurred in the canopy, with many species detected exclusively in that stratum, illustrating how terrestrial-only surveys fundamentally undersample rainforest primate assemblages. Furthermore, we utilized piece-wise exponential additive mixed models (PAMM) to provide rare evidence of cross-strata trophic dynamics, revealing that the native predator Cryptoprocta ferox exhibits a significantly higher and more targeted hunting pressure following Eulemur albifrons primate detections compared to terrestrial rodent prey. By documenting the first appearance of invasive Felis catus (a known predator of small primates) and the likely local extirpation of the native carnivore Eupleres goudotii, this work provides both a methodological roadmap for canopy-inclusive monitoring and urgent conservation data for one of Madagascar's most diverse primate habitats.

Image

Ph.D. student, Molecular Cancer Biology

Advisor: Ann Marie Pendergast  |  Collaborators: Jing Jin Gu

Role of ABL kinases in regulating DNA damage repair and transcription-replication conflicts in SCLC tumors

Small cell lung cancer (SCLC) is a highly aggressive neuroendocrine cancer that is typically metastatic upon diagnosis and has poor overall survival. Few therapies have yielded lasting effects, as most SCLC patients experience recurrence. Identifying novel molecular targets of SCLC is necessary for therapeutic development. We have found that inactivation of ABL tyrosine kinases impairs the outgrowth of metastatic SCLC tumors, resulting in prolonged animal survival. Treating SCLC cells with ABL allosteric inhibitors dysregulated DNA repair and DNA damage response pathways. ABL inactivation increases accumulation of transcription-replication conflicts (TRCs), compromises replication fork progression, and impairs the function of proteins implicated in transcription-coupled homologous recombination. To identify ABL interactors or substrates involved in DNA repair and genomic stability, we examined a previously generated ABL interactome dataset from our lab and identified several proteins with enriched binding to ABL. Among these ABL-interactors, DDX5, an RNA helicase known to resolve R-loops, emerged as a top target. DDX5 is highly expressed in SCLC and other tumors but lacks clinically approved drugs to target its function or stability. We found that ABL inactivation decreases DDX5 protein expression in SCLC cells. Ongoing studies will examine the role of ABL-mediated DDX5 activity in resolving R-loops and preventing TRC accumulation. Further, we will identify other targets that, when co-inhibited with ABL, exacerbate DNA damage induction and synergistically kills SCLC cells.

Image

Ph.D. student, Molecular Cancer Biology

Advisor: Ann Marie Pendergast  |  Collaborators: Roberto Barbier, Colleen Wu, Ann Marie Pendergast

Preventing Cell Migration in Triple Negative Breast Cancer

All cells have large communication networks that are constantly firing to help cells stay alive, grow, and move as needed. These communication networks, called signaling pathways, allow cells to adjust to new environments. Occasionally, cells may find themselves in an environment of limited oxygen, called hypoxia. When in hypoxic conditions, a protein called hypoxia inducible factor 1α (HIF-1α) begins to direct signaling pathways to help cells stay alive until oxygen is restored. Cancer cells are almost constantly under low oxygen conditions; therefore, HIF-1α is constantly directing various signaling pathways, promoting cancer cell survival. Because HIF-1α is more active in cancer cells than normal cells, we wanted to know if it had any unusual behaviors that could be exploited to prevent cancer spread or growth. 

Our work found that in triple negative breast cancer (TNBC), HIF-1α interacts with two other proteins, TAZ and TEAD4, which typically do not interact with HIF-1α. This group of proteins turns on a signaling pathway that promotes cancer cell migration. Migration, or metastasis, is what leads to the most deaths related to TNBC diagnosis. However, this group of proteins relies on the activity of another protein, called ABL. When we treat cells with an FDA-approved drug, Asciminib, it turns off ABL which disrupts the activity of the HIF-1α-TAZ-TEAD4 group to prevent cell migration from occurring. These findings highlight both the importance of hypoxia and HIF-1α in promoting cancer cell migration and reveal a method by which this progression can be stopped, improving patient care and survival.

Image

Ph.D. student, Biology

Advisor: François Lutzoni  |  Collaborators: Amber White, A. Elizabeth Arnold, François Lutzoni

Evaluation of Passive Aerial Spore Trapping Techniques

Fungal dispersal is a cryptic phenomenon with important outcomes for ecosystems and human health. Ranging from symbionts to saprotrophs, fungi perform essential functions in all terrestrial biomes of the Earth. Yet, questions regarding the dispersal of most fungi persist due to challenges associated with capturing and characterizing the microscopic spores used for fungal propagation. Previous studies have utilized passive spore trapping devices to capture deposition of aerial fungal communities. However, a lack of consensus on spore trap designs and sampling intensity limits the utility of these devices. Systematic comparisons between different spore trap designs are needed to identify the best practices for collecting passive deposition of fungal air spora. 

This study sought to compare the performance of three widely used passive spore trapping techniques: adhesive, liquid, and filter-based spore trapping. Twenty replicates of each spore trap were deployed within North Carolina’s Duke Forest to assess both inter-method variability and the environmental stochasticity of air spora, which contributes to variation between replicates. After one week of outdoor exposure, molecular methods of DNA extraction, PCR, and amplicon sequencing were used to characterize and compare fungal communities from spore trap samples. Species accumulation curves were also generated for each spore trapping technique to determine the level of replication needed to capture aerial fungal diversity. Altogether, this study provides novel insights into the methods used for capturing airborne fungal propagules, which will support future research to expand our narrow understanding of fungal dispersal.

Image

Ph.D. student, Marine Science and Conservation

Advisor: David Johnston  |  Collaborators: Brinnae Bent, David William Johnston

Can We Trust AI to See the Marine Mammals?

Artificial intelligence is increasingly helping scientists monitor marine mammals and understand changing ocean ecosystems. But when an AI system detects a whale, seal, or other marine animal, how do we know why it made that decision, and when should we trust it? 

Our interdisciplinary collaboration between the Duke Marine Robotics and Remote Sensing Lab and the Duke TRUST Lab explores how explainable AI (XAI) can make marine mammal detection systems more transparent and trustworthy. One approach we investigate uses generative AI to realistically “fill in” parts of an image, allowing us to explore how changes to an animal or its surroundings affect an AI detector’s decisions. Rather than treating AI as a black box, we can use these explanations to better understand its behavior and identify when it may be making unreliable predictions. 

At the research fair, we will make this work accessible through visual examples, an interactive activity such as a short quiz, and 3D-printed marine animals that invite visitors to explore whether and why we should trust an AI’s decision. This work highlights the value of bringing together marine science, robotics, conservation, and trustworthy AI. As AI becomes increasingly important in scientific research and environmental decision-making, building systems that people can understand, question, and trust will be essential not only for the future of marine conservation but also for interdisciplinary research at Duke and beyond.

Image

Ph.D. student, Physics

Advisor: Daniel Scolnic  |  Collaborators: David Jones, Richard Kessler

Calibrating the Cosmos: How Tiny Measurement Errors Could Shape the Next Century of Dark Energy Science

For 30 years, astronomers have known that the universe isn't just expanding—its expansion is speeding up. This accelerated expansion is driven by a mysterious "dark energy" that makes up about 70% of the universe yet remains almost entirely unexplained. 

To study it, we use a special kind of exploding star, the Type Ia supernova, as a cosmic mile-marker or ruler: because these explosions are all nearly identical in brightness, comparing how bright they look tells us how far away they are, and thus how the universe has grown over billions of years. Two enormous new telescopes—the Vera Rubin Observatory and the Nancy Grace Roman Space Telescope—are about to discover more than a million of these supernovae, hundreds of times more than we have today. But more data only helps if we can trust our measurements. Reading a supernova's distance depends on knowing our own instruments to extraordinary precision. 

My research asks a practical question: when those rulers are slightly off, where in our analysis does that tiny error do the most damage? Using detailed simulations of both telescopes, I traced how small calibration errors ripple through the whole pipeline. I found, contrary to what the field had assumed, which stage is most vulnerable, and that this dominant error can't simply be cleaned up later; it has to be controlled from the start. That result tells these billion-dollar telescopes exactly where to focus their effort to unlock the full story of dark energy.

Image

Ph.D. student, Environment

Advisor: Nishad Jayasundara  |  Collaborators: Kathryn Homa, Melissa Chernick, Jason Somarelli, Nishad Jayasundara

Integrating computer models into laboratory studies to understand the chemical problem in plastics

Plastics are everywhere and are a source for thousands of chemicals in our everyday lives and in the environment. These chemicals are used in the production of plastics to improve their properties and impart various functions. They fall into various classes including plasticizers, anticorrosives, antioxidants, UV-stabilizers, flame retardants, surfactants, and cross-linking agents. 

We aim to investigate the consequences resulting from exposure to these chemicals using bioinformatics that are validated using a laboratory animal model. Existing databases combine data from thousands of studies that used to analyze the health impacts of chemicals of interest using bioinformatics. 

Our data suggest chemicals in plastic additive target mitochondrial end points. Zebrafish are a valuable model for both the study of human disease and environmental toxicology. We quantified various parameters related to mitochondrial function resulting from these chemical exposures. Our results showed effects in 30 of 39 chemicals including organophosphates, phthalates, bisphenols, polycyclic aromatic hydrocarbons, parabens, per- and polyfluoroalkyl substances, industrial amides, UV-filters, and aromatic amines. The mitochondrial function assay provided a powerful tool for plastic additives screening, given the emergence of mitochondrial health as a key indicator of subcellular health that tracks of human disease susceptibility as well as ecological fitness of aquatic organisms. Going forward, this provides a framework for emerging chemicals and identifying safer alternatives from both public health and regulatory perspectives.

Image

M.S. student, Statistical Science

Advisor: Vahid Tarokh  |  Collaborators: Benjamin Doshna, Junyi Liao, Vahid Tarokh, Simon Sponberg

The Hidden Signals Behind Moth Flight

All movements we make are from the electrical signals sending out from our brain to muscles, and in humans this process is complicated and involves millions of neurons and hundreds of muscles, which makes it hard to study. However, hawkmoths give us the chance to simplify the problem by using only ten muscles to power their movement and chase the flower. Those signals from neurons to muscles are invisible but can be heard by electrodes. 

We use the robotic flower with unpredictable movements as its target, and measure the forces moths generated in different directions roughly twenty-five times per second. Then we use machine learning tools to study the spike patterns and their hidden structure related to movements. Surprisingly, those patterns can be compressed to essentially only one number per wingbeat, and the number is closely related to the produced forces and torques that are used for movement. Even after they are fed and heavier, the command does not seem to change direction and keeps the same steering strategy. Overall, the recorded moth muscle spikes are a rare dataset, and it gives us a complete view of how the nervous system controls motion using those spikes. By studying the moth, we can gain a fuller understanding of how the motor program works biologically and potentially apply these ideas to future technologies like neural prostheses that benefit human lives.

Image

Ph.D. student, Biomedical Engineering

Advisor: Timothy Dunn  |  Collaborators: Sihan Lyu, Lauren Lim, Yuxuan Wen, Timothy Dunn

From a Score to a Measurement: Objective Movement Assessment for Parkinson's Disease with TULIP

Parkinson's disease (PD) affects more than 10 million people worldwide and about 1.1 million in the United States. Someone in this country is diagnosed every six minutes, and the count rises as the population ages. PD directly affects the quality of life, through symptoms such as tremor and slowed movement. However, the way we measure its severity is still subjective. When you go to the clinic, the clinician will ask you to perform activities, including walking and rotating your forearms. Based on your movement, they assign a score from 0 (healthy) to 4 (severe). This ordinal score system invites disagreement across clinicians, which makes progression tracking imprecise worldwide. 

To solve this problem, we design a markerless multi-camera system to capture the body, face, and hands accurately and quantitatively. Unlike the existing scales, our system can track detailed and explainable movement features of patients. For example, instead of a score-one gait representing mild severity, our system captures the subject's step width in centimeters, or leg angle changes in degrees while walking. We named this system TULIP (Three-dimensional Understanding and Learning of Impairment in Parkinson's disease), and the red tulip is the symbol of PD. The system is installed in the Duke Neurology Clinic and has recorded more than a hundred participants performing hand movements, gait, and everyday activities such as drinking. In our recent works, these explainable movement measurements reproduced standard motor assessment and estimated disease duration. We now target integration with medical records, towards personalized medicine.

Image

Ph.D. student, Biomedical Engineering

Advisor: Shyni Varghese  |  Collaborators: Jiaul Hoque, Shyni Varghese

One Material, Two Jobs: A New Approach to Osteoarthritis

Osteoarthritis affects hundreds of millions of people, yet most treatments primarily manage pain rather than addressing the damages occurring inside the joint. Our research asks, Can one injectable material do two jobs at once, helping an arthritic joint move more smoothly while also delivering a therapy that protects cartilage? We are developing microsized particles called microgels made from hyaluronic acid, a material naturally found in joints. We modify their surfaces to improve lubrication and reduce friction against cartilage. Inside these microgels, we incorporate small sugar-based carriers that stabilize FGF-18, a protein that promotes cartilage repair but can lose activity during storage and delivery. Our engineered microgels reduce friction against cartilage and help preserve FGF-18 activity. In a mouse model of post-traumatic osteoarthritis, a single injection of FGF-18-loaded microgels reduced pain-related behavior, cartilage degeneration, and joint inflammation compared with untreated injured joints. 

This work illustrates a broader idea in biomaterials engineering: instead of designing a material simply to carry a drug, we can engineer the material itself to perform therapeutic functions. By addressing both the mechanical and biological aspects of osteoarthritis, multifunctional materials like these could help inspire new approaches to treating joint disease.

Image

Ph.D. student, Electrical & Computer Engineering

Advisor: Aaron Franklin

Printing the Way You See the World: The Future of Flexible Displays

How much of your day do you spend looking at a screen? From phones and laptops to televisions, cars, and wearable devices, displays have become an integral part of how we experience the world. We spend hours every day looking at them and yet, the displays themselves have changed surprisingly little in how they are made. They are still largely rigid, flat, and built through conventional manufacturing processes. 

In our lab, we develop technologies that could enable displays to become thinner, and more adaptable to the surfaces around us. A central challenge is how can we make flexible displays respond quickly enough for the demanding applications of tomorrow? My work addresses this through carbon-nanotube thin-film transistors (CNT-TFTs) which are tiny printed switches that control how light and information flow through a display. I investigate these challenges at a fundamental level i.e., how many pixels can we actually fit into a printed display before the pixels start interfering with each other[1], and how can we control the brightness of a display using better light-reflecting nanomaterials like quantum dots (QDs)? Ultimately, my goal is to move displays beyond rigid screens toward thin, flexible, and potentially transparent surfaces that integrate seamlessly into our surroundings; from next-generation televisions and wearable technologies to the interiors and windows of future vehicles. Rather than asking where we will look at screens next, my research asks: what if the screen could be anywhere? 

[1] Width-Spreading Effect in Printed Carbon Nanotube Thin-Film Transistors and Its Impact on Device Performance. ACS Appl. Nano Mater. 15 May 2026; 9 (19): 8781–8788. https://doi.org/10.1021/acsanm.6c00640

Image

M.Eng student, AI for Product Innovation (Pratt)

Advisor: Brinnae Bent  |  Collaborators: Tiffany Degbotse, Sharmil Nanjappa, Jiayi Zhou, Vishnu Manoj, Vihaan Nama, Boyuan Chen, Lee Tiedrich, Christopher Bail, David W. Johnston, Walter Sinnott-Armstrong, Brinnae Bent

Aïra: Rethinking AI Research Assistants for Interdisciplinary Science

Solving today’s complex problems requires researchers from different domains working together. However, interdisciplinary collaboration introduces challenges that make it difficult for the team to work on a shared understanding of the problem, including differing vocabularies, perspectives, and conceptual frameworks. While current AI research assistants serve as valuable tools for supporting an individual researcher by answering questions and providing documents or assistance for an user with only one perspective, current AI systems are not designed to support interdisciplinary teams and collaborative reasoning. 

We propose aïra; an AI Interdisciplinary Research Assistant designed to support interdisciplinary scientific collaboration. Aïra transforms recorded discussions from interdisciplinary meetings into structured collaboration frameworks. Rather than generating a conventional meeting summary for one perspective, aïra analyzes the discussion through the lens of interdisciplinary collaboration. The assistant identifies each perspective in the conversation, priorities and next steps, decisions with their discussed tradeoffs, assumptions and their impact, differing terminology, and missing disciplines that should be involved in future discussions before synthesizing these into a shared multi-domain representation of the interdisciplinary meeting. The system utilizes four agents to extract information, recognize gaps through a fine-tuned LLM serving as a domain expert, understand decision rationale, and compile a vocabulary. Aïra can be found at researchwithaira.com. Aïra is designed to help research teams align in collaborative reasoning and recognize lost conceptual connections across projects that span months or years. We hope this work encourages a focus in the development of AI research assistants that strengthen communication, integration, and discovery across disciplines.

Image

Ph.D. student, Marine Science and Conservation

Advisor: Daniel Rittschof  | Collaborators: Corin Hoppe, Opal Mohs, Matthew Casavant, Josh Matheson, Beatriz Orihuela, Gary Dickinson, Laura Dalton, Aicha El Kharr, Mark Wiesner, Anselme Dossou, Dan Richter, Dan Rittschof

3D-Printed Plastics Have Cryptic Consequences on Marine Animals

Materials can be durable, flexible, biodegradable, and increasingly, 3D-printable. Those same properties may also affect the organisms that encounter them in the ocean. Because comprehensive safety testing is too slow and expensive to apply to every promising material, marine science needs practical early-warning systems. 

This research uses barnacles as one such system. Barnacles attach to hard surfaces and build calcium-carbonate shells, which is the same mineral foundation used by corals and oysters. They therefore provide a rapid model for studying how materials may affect other ecologically and economically important calcifying organisms. Barnacles were grown on 3D-printed PLA, ABS, nylon, and TPU and compared with barnacles grown on glass, a biologically neutral reference surface. Weekly external morphometric measurements showed that growth trajectories differed among plastic substrates. High-resolution microcomputed tomography then provided the study’s “X-ray vision,” creating three-dimensional images of shell structure without breaking the animal. This revealed internal changes in shell volume, thickness, shape, and symmetry that external measurements can miss. Barnacles grown on PLA had significantly thinner, lower-volume shells than those grown on glass, despite appearing similar in overall size and shape. These hidden changes may influence the structural foundation through which calcifying animals attach, withstand waves, and persist in marine environments. By revealing material effects early, barnacles can replace assumptions with evidence before new 3D-printed materials are used at scale in the ocean.

Image

Ph.D. student, Biomedical Engineering

Advisor: Timothy Dunn  |  Collaborators: Alexa Izquierdo, Sophie Shi, Joshua Wu, Tim Dunn

Detecting emotions in freely moving animals

Measuring an emotional response in animals that can not verbally articulate presents a challenge for forming a rigorous understanding of emotions across species. While invasive methods can be used to record emotional signals directly from the brain, having a completely non-invasive method reduces stress on the animals. Using high-speed camera recordings and careful tracking of body and body-part movements of freely-moving mice, my research seeks to detect emotional responses non-invasively.

Image

M.S. student, Computer Science

Advisor: Bhuwan Dhingra  |  Collaborators: Bhuwan Dhingra, Marcos Zampieri, Nishat Raihan, Dipak Meher

What does homework mean in the age of AI?

Generative AI, especially Large Language Models (LLMs), has fundamentally disrupted traditional university education. While LLMs may offer powerful learning support, students can also use them to bypass genuine effort by generating entire essays or programming code, and obtaining answers to conceptual questions. This forces instructors to ask a critical question: If AI can do the homework, how do we know what students actually know? This challenge is especially pertinent in technical subjects like Computer Science which rely on artifact-centered, unproctored tasks like coding projects that can be convincingly completed with AI assistance. Put simply, the foundational skills and knowledge of the next generation of innovators are at stake. 

To study this phenomenon, we conducted a systematic review of 135 papers addressing CS assessment in this era of generative AI. We analyzed existing assessment practices, the forms of evidence they produce, and the ways GenAI weakens this evidence, alongside new ideas emerging in the literature. Our research led us to a core "assessment trilemma"—the struggle for assessment methods to be (1) authentic to modern practice, (2) scalable for large courses, and (3) trustworthy as evidence of learning at the same time. Finally, we map the tradeoffs of emerging solutions to guide educators through this evolving landscape, with the ultimate goal of developing robust tools and methodologies that allow modern educators to maximize student understanding.

Image

Ph.D. student, Environment

Advisor: Susan K. Murphy

Leveraging Your Cells to Understand Your Environment

Using cells derived from humans (induced pluripotent stem cells or iPSCs) to grow thousands of mini-brains in a dish. These mini-brains are then exposed to chemicals, including cannabis and heavy metals to understand how they affect early fetal neurodevelopment. Therefore using a human-to-dish conversion (bypassing the need for animal experiments) to get safe-levels of consumption and exposure to compounds that are widely prevalent.

Image

M.S. student, Statistical Science

Advisors: Megan Kelly-Deyncourt / Dawn M. Strickland  |  Collaborators: Boshuo Wang

Making Brain Stimulation Therapy Faster and Safer Through Smarter Stopping Rules

Transcranial magnetic stimulation (TMS) uses magnetic pulses to stimulate specific brain regions and treat conditions such as depression. Before treatment, clinicians determine each patient's motor threshold—the stimulation intensity needed to reliably produce a motor response. This threshold is used to calibrate treatment dosage, making accurate measurement important for effectiveness, reproducibility, and safety. Motor responses vary naturally from pulse to pulse, making threshold determination difficult. 

Our lab developed SAMT, a motor-thresholding procedure that uses stochastic approximation. SAMT adjusts stimulation intensity as responses are collected, without requiring an assumed mathematical model of how a patient will respond. With an appropriate starting intensity, it can determine motor threshold efficiently and accurately. However, deciding when a SAMT session has collected enough evidence remains challenging. The current app provides an approximate quality indicator, but it may declare an estimate reliable too early; therefore, the operator must make the final judgment. Some sessions may also stabilize earlier than expected, making additional pulses unnecessary. 

My research develops a per-pulse convergence classifier and dynamic stopping policy for SAMT. After each pulse, the classifier evaluates the accumulating session history and estimates whether the running threshold estimate is sufficiently accurate. Stopping requires sustained confidence across consecutive pulses, helping distinguish stable convergence from brief periods of apparent accuracy. The policy is designed to limit premature stops to 2%, verified on data withheld from model development, while reducing unnecessary stimulation and session time.

Image

Ph.D. student, Public Policy

Advisor: Lisa A. Gennetian

Spillovers of Old Age Allowances on Children's Health: Evidence from Bangladesh

Studies of government-funded allowances for the elderly find positive spillover effects on other members of the eligible recipient's household. In multigenerational households, transfers to the elderly may affect child health through improved food security, greater health care utilization, changes in parental labor allocation, or changes in influence over household spending decisions. Yet how such transfers affect child nutrition remains empirically poorly understood, especially in rural South Asia, where older adults and young children often share household resources. 

Bangladesh's Old Age Allowance (OAA) program, an unconditional cash transfer to the poor elderly, is the country's largest means-tested program and offers an important setting to examine whether means-tested cash transfers produce intergenerational benefits. Three preliminary findings emerge. First, we find a reduction in child mortality, albeit imprecise, and no meaningful relationship with fertility. Second, we find exposure during ages 0-2 is associated with a lower prevalence of a child being underweight. However, we find no impact on wasting and a higher, though imprecise, prevalence of stunting. Third, we examine older children and find that OAA exposure is associated with a lower prevalence of being underweight and stunted among children aged 5 or older. Findings contribute to our understanding of how social protection for the elderly may benefit children living in the same household, while also speaking to broader questions about the intergenerational effects of cash transfers, household resource allocation, and child well-being in low-income rural settings.

Image

Ph.D. student, Psychology & Neuroscience

Advisor: Maureen Craig 

Black Americans Prioritize White Americans’ Values Over Their Economic Resources as Allies

Psychological research has identified several characteristics that predict Black Americans’ perceptions of advantaged group members as trustworthy allies, including having the same views on race and racism as oneself, having genuine motivations to reduce inequality, and focusing on empowering marginalized groups. Sociological research has also revealed important characteristics: an ally’s ability to mobilize others or financially invest in causes. Yet, research often focuses on views of White American allies varying on one characteristic at a time. It is not yet clear how these characteristics are perceived within racial minority allies and which characteristics can be most impactful in reducing racial inequality. 

Three studies (N = 595) use a highly-powered preregistered conjoint design to simultaneously test what ally characteristics Black Americans prioritize. We find that Black Americans chose White, Asian, and Hispanic American allies with high racial shared reality, genuine motivations, and centering actions, and put a secondary priority on social and economic resources. This work provides insight into how Black Americans view allyship characteristics across diverse allies and has implications for how to build trust between different social groups. 

Image

Ph.D. student, Psychology & Neuroscience

Advisor: Jennifer Groh  |  Collaborators: Shawn M. Willett, Nicholas Marco, Surya T. Tokdar, Jennifer M. Groh

How the brain processes multiple sounds

In everyday life, we are constantly bombarded by many different sounds. However, for most, our brains can clearly parse these sounds so that we can tune in to our friend’s voice in a crowd or the violin part within a symphony. 

My research investigates this process. Particularly, the current work focuses on how the brain maintains information about more than one sound. We found that in response to two simultaneous sounds, individual neurons can switch back and forth over time between encoding each signal. Further, we found that this switching behavior is more likely when the two neural responses to each individual sound are more similar. This means that neurons switch back and forth to distinguish between two signals that are otherwise difficult to separate. This research has the potential to inform investigations of a variety of hearing and brain disorders. It is particularly relevant to understanding why some individuals have difficulty understanding sound (including speech) in a noisy environment. Elucidating the code used by the brain to represent multiple sensory objects could also be used to improve and develop new sensory prosthetics (including brain-computer interfaces).

Image

M.S. student, Global Health

Advisor: Maegan Hoss  |  Collaborators: Angella Lagat, David Khaoya, Adams Akullian

Where First, To Whom, and At What Cost? Targeted Lenacapavir Distribution for HIV Prevention in Kenya

Kenya has approximately 1.4 million people living with HIV, with 20,000 new infections annually, despite the existence of effective HIV prevention tools like Pre-exposure Prophylaxis (PreP). Lenacapavir, a revolutionary medication approved by WHO (2025) is a single-dose injection every 6 months that prevents HIV with 99% efficacy overcoming many limitations of daily oral PrEP by removing the need for daily pill-taking and reducing adherence-related failures (WHO, 2025). However, high costs and limited supply require strategic deployment. The Global fund donated 1,000,000 doses and a $100 million budget to Ministry of Health (MoH), Kenya. But with 47 counties and millions of people at risk, there is not enough for everyone. The critical question becomes: where should the doses go first—and to whom—to achieve the highest impact and cost-effectiveness? 

Working with Centre for Epidemiological Modelling and Analysis, University of Nairobi, we built a mathematical model comparing four geographic distribution strategies; covering 30%, 60%, 90%, and 100% of the target population across counties in Kenya. Using county-level HIV data from UNAIDS, the model measured which strategy prevented the most infections at the lowest cost. The findings were striking. Targeting 90% coverage in high-incidence counties prevented 2,637 HIV infections and 52,740 years of healthy life lost at a cost of $1,896 per year of healthy life saved. Spreading doses uniformly across all counties was both less effective and less efficient. The largest gains occurred between 30% and 60% coverage, while returns beyond 90% were minimal making 90% the strategic sweet spot.

Image

Ph.D. student, Psychology & Neuroscience

Advisor: Maureen Alyson Craig

Unequal Odds: How Social Hierarchies Shape Perceptions of Lopsided Competition

Research on zero-sum competition, an extreme form of competition where one’s gains are seen as another’s loss, has mostly examined competing groups’ perceptions of each other. This has left uninvolved third-party perceptions understudied despite its importance in contexts that involve multiple social groups. As hierarchies may comprise multiple social groups, it is a unique context within which we may observe such third-party perceptions of competition between two social outgroups—groups in which third-parties do not hold membership. 

We conducted three studies (N = 724) to examine Americans’ perceptions of zero-sum competition within two social hierarchies, the diverse U.S. racial hierarchy and a fictitious international economic hierarchy. Studies 1 and 2 revealed that Black Americans reported that the socially-dominant group’s (White Americans') gains came at a marginalized group’s (Asian Americans') expense more so than vice versa, but not when shown information of the latter having greater resources than the dominant group. Participants in Study 3 learned about two fictitious countries and their economic conditions. After viewing information about an unequal economic hierarchy wherein one country had greater resources than the other, participants reported beliefs that the high-status country’s gains came at the low-status country’s expense more so than vice versa, but no such asymmetry appeared before participants had this information. Together, we find that in highly unequal hierarchies, third-parties perceive competition as lopsided contests in favor of high- over low-status groups; without it, they do not perceive predetermined winners and losers, thus advancing work on intergroup dynamics in diverse societies.

Image

Ph.D. student, History

Advisors: James Chappel and Hannah Conway

Space Before the Race: How United States Popular Culture Imagined Human Space Travel from 1920-1955

Almost everyone seems to know about the popular culture that came from the Space Race of the twentieth century: shows like Star Trek, The Jetsons, Doctor Who and movies such as Star Wars, 2001: A Space Odyssey, and Close Encounters of the Third Kind remain cultural touchstones to this day. Space historians have spent the last few decades outlining the importance of this popular culture in the development of the geopolitical Space Race competition between the United States and the Soviet Union. Now, however, it is time for space history to enter a new frontier. 

My research does just that by asking what did space-themed popular culture look like before the Space Race? How were American writers, entertainers, and everyday Americans imagining and/or told to imagine space and the American place within it both in the contemporary moment and in imagined futures? How did this content reflect other contemporary issues Americans were facing? Looking at books, comics, movies, television shows, and more, my dissertation project hopes to highlight that U.S. space history has a larger story to tell outside of the Space Race and that story might just hold the answers to many of the questions we are still grappling with today as we once again look towards the stars.

Image

M.A. student, Teaching

Advisor: Christina Gibson-Davis  |  Collaborators: Dr. Christina Gibson-Davis, Dr. Portia Miller, Dr. Elizabeth Votruba-Drzal, Dr. Jaime Hanson, Teresa Longnecker, Sheri Petrequin, Amber Liller, Abby Sites, Hannah Parker, Terrique Morris, Darin Fields

The Study of Teen Experiences that Promote Success (STEPS)

The goal of our research is to understand how family economic circumstances affect adolescent functioning. Through quarterly surveys over the course of two years with one teen, enrolled between ages 13 and 15 and one caregiver, STEPS seeks to acquire longitudinal data into the mental and economic well-being of Durham, NC and Pittsburgh, PA area teens. Through this work, we hope to better understand how family economic circumstances affect youth functioning, behavior, health, and future goal-setting.

Image

Ph.D. student, Psychology & Neuroscience

Advisor: Mark Zachary Rosenthal  |  Collaborators: Bill Chen, Ashley A. Moskovich, Marta Siepsiak, Siyuan Wang, M. Zachary Rosenthal

When Everyday Sounds Feel Unbearable: Tracking Misophonia in Real Time

Why can ordinary sounds like chewing, breathing, or tapping trigger intense distress in some people? Misophonia is a condition in which specific everyday sounds can evoke unusually strong emotional reactions, yet we still know relatively little about how those reactions unfold in daily life. 

In this study, we used real-time smartphone surveys to follow 40 adults with misophonia as they encountered trigger sounds in their everyday environments. Negative emotion rose sharply following a trigger and often remained elevated for 30–60 minutes before moving back toward baseline. We also found that people reporting greater fear at baseline tended to experience stronger emotional reactions to later triggers. At the research fair, visitors will be able to follow a typical trigger episode—from hearing an everyday sound through the emotional response and recovery that follows—and guess how long they think the reaction lasts before seeing the study results. Simple visuals and familiar sound examples will make the findings easy to explore without a psychology background. Understanding not just whether people react to trigger sounds, but how their emotions change and recover over time, may provide clues about when emotion-regulation strategies could be most useful. More broadly, this work can help us understand how emotional responses to distressing experiences unfold in everyday life.

Image

Ph.D. student, Psychology & Neuroscience

Advisor: Jennifer Groh

How The Brain Processes Multiple Sounds

My research is on how the brain processes multiple sounds at once! Although it may sound like a trivial question, and is something we do seamlessly every day, the neuroscience of how it happens is actually pretty complex. Neurons represent information in the brain through firing rate. So if a stimulus comes on, the neuron responds with a series of action potentials, or spikes, which represent that stimulus. Different stimuli are represented by different firing rates; a neuron might fire 80 spikes/second to represent a red light, but only 40 spikes/second for a yellow light. 

The question my research seeks to answer is what happens when two stimuli are presented at the same time. Mathematically, the neuron can't just average the two firing rates each stimulus would elicit, because there would be a lot of information loss. To study this, I use macaque monkeys, and I record the electrical signals directly from their brain in a sound-processing region called the auditory cortex, as they listen to a variety of sounds. It turns out the answer is in timing. When presented with both a sound A and a sound B at the same time, the neuron will fluctuate in time between responding to one vs the other. So on some trials it will fire with a rate A, and on other trials with rate B, and we can statistically analyze the firing patterns to reveal how the neuron responds to both. This is a powerful theory for not just how the brain represents not just multiple sounds, but any combination of stimuli, and provides insight into how we may better understand hearing loss and other sensory disorders.

Image

M.A. student, Applied Ethics and Policy

Advisors: David A Hoffman (Program), Marc Groman (Career)  |  Co-Presenter: Moonwon Seo 

National Narratives and Modern AI Policy: An alternative angle of the "AI race"

People say there's an AI race to win. We don't think so. Drawing from historical policy documents, leadership speeches and government-led initiatives from the U.S., China and EU, we show that there is more to technological policymaking than material interests and immediate gains. In fact, each country follows a clear narrative trajectory that can be traced through the ages. The U.S. wants to be the first to uncover new frontiers; China wants to make sure its matters are firmly in its own hands; and the EU, born out of post-WWII trauma, is adamant on preserving their market economy and value-based politics. These stories persist even now and have greatly influenced why each country’s AI governance regime has taken its shape today. AI safety is an issue that requires coordination across national borders. A zero-sum framing focusing on the “AI race” makes such coordination appear impossible. However, by understanding AI policy as products of narratives rather than standalone events, we reveal both the inefficiencies of current multilateral efforts, and where latent points of convergence exist. In fact, all three countries, despite their differences, have already converged on issues of mutual interest like content provenance, serving as evidence that international collaboration on AI safety is both possible, and already happening.

Image

Ph.D. student, Public Policy

Advisor: Matthew Johnson

Safeguarding Labor Rights: The Impact of Worker Centers in the United States

Worker centers are hybrid community-based organizations that support workers through a mix of direct service provision and political advocacy focused on labor and immigrant rights. They have become an important labor market institution in the U.S. in recent years. This paper examines the causal effects of worker centers on workplace injuries, complaints, programmed inspections conducted by the Occupational Safety and Health Administration (OSHA), and labor compliance among nearby establishments. By exploiting the staggered rollout of the opening time of worker centers across ZIP codes, I find that worker centers result in a significant decline in workplace injuries by 4.47%, a marginally significant increase in programmed inspections by 6.44%, a significant decline in the average number of violations per programmed inspection by 10% to 14%, and an insignificant increase in worker complaints. I also find that the effects are highly localized, with significant effects concentrated within five miles of worker centers. By providing novel empirical evidence, this paper has significant implications for policymakers and scholars concerned with labor rights, civil organizations, and the future of labor regulations in the U.S.