2026 CompX Grantees
About the CompX Faculty Grants Program
Winners of the 2026–2027 Neukom Institute CompX Faculty Grants Program at Dartmouth have been announced for one-year projects, marking the largest application cycle in the program's history. The competition drew over $1.5 million in total funding requests. In response, the Institute awarded a combination of direct funding, Neukom Scholar research assistant support, and research computing resources totaling $490,000.
The program is designed to support both the development of novel computational techniques and the application of computational methods to research initiatives across Dartmouth's campus and professional schools.
Faculty from across Dartmouth College—including undergraduate, graduate, and professional schools—were eligible to apply for these highly competitive grants.
Anthropology: Jesse Casana
Advancing 3D Ground Penetrating Radar for Archaeology
As conventional archaeological excavations face growing ethical, logistical, and preservation challenges, an emerging suite of geophysical technologies is enabling researchers to document subsurface remains without disturbing the ground. New ground-penetrating radar (GPR) systems represent a particularly powerful advancement, integrating multi-channel, step-frequency antenna arrays, GPS-enabled data collection, and advanced imaging software to produce exceptionally high-resolution, real-time 3D imagery of the subsurface.
This project will acquire a state-of-the-art GPR system and deploy it in surveys of archaeological sites across Maine, Oman, Tunisia, and Arizona. These efforts—spanning the largest application cycle in the program's history—will generate novel insights into the development of ancient settlements worldwide, while also supporting hands-on student training and advancing innovative, non-invasive methods for exploring the human past.
Thayer School of Engineering: Peter Chin
Microwave-Based, Multi-Modal Signal Processing and Foundation Model Framework for Non-Invasive Detection of Sarcopenia
Sarcopenia—the pathological infiltration of fat into muscle tissue—leads to significant muscle weakness, reduced mobility, and markedly poorer outcomes following surgery, chemotherapy, and radiation therapy. The current gold standard for diagnosis is X-ray computed tomography (CT); however, the high radiation dose limits its viability as a routine screening modality.
Microwave interrogation presents a compelling alternative for non-invasive sarcopenia detection. The technique leverages the stark contrast in dielectric properties between muscle and fat: across the 1–8 GHz frequency range, muscle exhibits permittivity values generally above 45, while fat is typically around 5. This near order-of-magnitude difference creates a strong signal basis for distinguishing tissue composition.
In this project, we have developed a pair of transmission probes using advanced 3D metal printing technologies. Data collected from these probes will be analyzed through three complementary signal-processing frameworks: raw frequency-domain spectra, wavelet decompositions that capture localized spectral features, and time-domain impulse responses.
We will also evaluate the use of time-series foundation models—such as TimesFM and Chronos—as feature extractors for sarcopenia characterization. The time-domain representation is particularly well-suited to these pretrained models, which are optimized for temporal pattern recognition. By applying these models in a zero-shot capacity, we will extract learned embeddings and train lightweight classifiers, avoiding the need for extensive fine-tuning. This approach directly addresses the small-sample constraints typical of pilot datasets, while leveraging robust representations derived from large-scale pretraining.
Classics: Rachel Dubit
Poetic Object Extraction Tool (POET)
How do formal phonetic and aesthetic features shape the emotional impact of a poem composed thousands of years ago? For centuries, Classics scholars have relied on manual counting methods to quantify stylistic features across texts ranging from the Homeric epics to Latin love elegy. This project brings modern computational approaches to these enduring questions through the development of an open-source web application: the Poetic Object Extraction Tool (POET).
With POET, researchers will be able to upload Greek poetic texts and receive detailed, automated analyses of key stylistic features, including hiatus, anaphora, anadiplosis, alliteration, and assonance. The platform will generate precise reports on the frequency, variation, and distribution of these elements, significantly accelerating and standardizing what has traditionally been a labor-intensive process.
With support from the Neukom Institute, the project will also expand into more nuanced linguistic territory by developing methods to analyze less easily quantified features, such as so-called "harsh" and "smooth" sounds. By applying neural networks and other computational techniques in this low-resource linguistic setting, the project aims to open new analytical pathways for researchers, educators, and students working with ancient texts.
The team includes Drs. Rachel E. Dubit and Maria Gaki, along with research associates Atanas Iliev, Luke LoTempio, and Felipe Leao (all Dartmouth '26).
Music: Ash Fure
Adaptive Sonic Intensity: Sensor-Based Environmental Awareness and Modular Resonant Architecture
This project develops adaptive sound environments capable of sensing and responding to the spaces and bodies around them. Addressing the ways museums and galleries often diminish immersive audio experiences, it proposes computational systems that preserve sonic intensity while remaining spatially and socially responsive.
The work is organized around three core research strands: embedding sensors into large-scale installations to dynamically modulate sound and light in real time; creating audience-facing interfaces that allow participants to calibrate their own level of sensory intensity; and designing modular resonant pods that function as acoustically responsive architectural instruments.
Together, these innovations position artistic practice as a rigorous experimental platform for advancing responsive environments, smart architecture, and embodied human–computer interaction.
Earth Sciences: Robert Hawley & Jeff Kerby
EPOCHS: Remote Sensing and Computer Vision for Tracking Tundra Wildlife Impacts, Year-Round
Large Arctic herbivores such as caribou and muskoxen play a critical role in shaping tundra ecosystems through grazing and trampling, yet their cumulative landscape-scale impacts remain difficult to quantify with existing monitoring tools. EPOCHS (Extended Polar Open-source Computer vision for Habitat Surveys) addresses this gap by integrating short-interval, solar-powered timelapse camera systems with AI-based animal detection and 3D terrain reconstruction to continuously track herbivore movement and associated surface disturbance across seasonal cycles. The platform generates spatiotemporal maps of animal activity, supporting ecosystem sampling design and enabling non-invasive population and behavior monitoring in remote and environmentally sensitive regions.
A central challenge is engineering a system that is low-cost, energy-efficient, and sufficiently rugged for year-round deployment in extreme polar conditions. Leveraging interdisciplinary expertise from Dartmouth's Glaciology Lab, the Institute of Arctic Studies, and the Citrin Family GIS/Applied Spatial Analysis Lab, the project will conduct iterative hardware and software field trials in both the Upper Valley and Arctic environments. The goal is to advance a scalable, autonomous monitoring system capable of capturing gradual ecological change in high-latitude landscapes and other extreme environments globally.
Studio Art: Karol Kawiaka
Unveiling Palladio's Design Method: High-Precision 3D Modeling of Roman Baths and Il Redentore
Andrea Palladio's final Renaissance masterpiece, the Church of Il Redentore, is widely believed to reflect his early studies of Roman bath architecture, although this connection has not been quantitatively demonstrated. This project will address that gap by developing detailed 3D digital reconstructions of the Great Halls of key Roman baths—including those of Agrippa, Caracalla, Constantine, Diocletian, Trajan, Hadrian, Titus, and Nero—based on Palladio's own drawings.
The research will combine archival investigation at the RIBA in London with on-site documentation of Roman bath structures. These sources will inform a suite of digital and physical outputs, including an interactive VR platform and the fabrication of 3D-printed and CNC-milled prototypes for exhibition purposes.
The final output will provide a rigorous proportional analysis of Palladio's design logic, testing the hypothesis that he employed literal, scalable modeling techniques derived from Roman precedents. In doing so, the project will offer scholars, educators, architects, and the public new tools—both virtual and physical—for engaging with Palladio's design process.
African and African American Studies: Tricia Keaton & John Bell
Black Paris Narrative
by Gia Kim
The Black Paris app leverages AI and locative technologies to enhance International Study Abroad (ISA) learning experiences. Combining Niantic's augmented reality platform with an on-device large language model, the mobile application uses GPS data and AI-driven visual analysis of student-captured photos to identify historically significant sites in real time.
The platform then introduces a novel "time-shifted" learning model, delivering curated cultural and historical content after site visits through delayed push notifications and Socratic-style AI chatbot conversations. This approach encourages reflection and deeper engagement beyond the immediacy of the physical experience.
Piloted during the Black Paris ISA program in summer 2027 with 20 undergraduate students, the project evaluates how AI-mediated, location-aware technologies can surface overlooked heritage narratives and meaningfully deepen student engagement.
Chemistry: Katherine Mirica
Computationally Guided Design of Conductive MOF Sensor Arrays for Selective Detection of Bacterial Volatile Metabolites
This project develops a computationally guided framework for engineering conductive metal–organic framework (MOF) sensor arrays that selectively detect volatile organic compounds (VOCs) generated by bacterial metabolism. Rapid identification of pathogen-specific VOC signatures has the potential to enable timely, targeted therapies while reducing dependence on broad-spectrum antibiotics that contribute to antimicrobial resistance. The work integrates materials design, chemical sensing, and machine learning to create a scalable platform for infection detection. Expected outcomes include quantitative structure–response relationships and a generalizable design strategy for selectively responsive chemical sensor arrays.
Physics & Astronomy, Engineering: Chandrasekhar Ramanathan & Peter Chin
Controlling Quantum Dynamics with Artificial Intelligence
Characterizing and precisely controlling quantum system dynamics is essential for advancing quantum technologies, including sensing, simulation, and computation. However, designing the required classical control sequences constitutes a high-dimensional, constrained optimization problem that is computationally intractable using conventional methods.
This CompX project proposes the use of agentic AI to autonomously discover and optimize control protocols for interacting spin systems in solids, with a focus on suppressing inter-spin interactions. Achieving such decoupling is critical for narrowing spectral linewidths in electron paramagnetic resonance (EPR) and nuclear magnetic resonance (NMR) spectroscopy, and for preserving coherence in solid-state, spin-based quantum information platforms.
Film and Media Studies: Roopika Risam
Can the Algorithm Write Back? Relational Epistemic Data for Reconfiguration through Experimental Situated Systems (REDRESS)
This project investigates how large language models (LLMs) encode and reproduce representations of human experience—and how those representations can be systematically reshaped through targeted interventions in training data. Bringing together an interdisciplinary team of scholars and students, the project will develop REDRESS, a fine-tuned, lightweight open language model designed to reconfigure narrative generation using insights from postcolonial and Black diaspora literature.
Central to the work is the design and evaluation of narrative constraints that shift perspective, agency, and voice in model outputs, enabling more nuanced and equitable forms of storytelling. By integrating computational methods with critical literary frameworks, the project advances both technical and humanistic approaches to AI development.
Outcomes will include an open-access model, a public-facing report, and scholarly publications that offer new frameworks for understanding human subjectivity, representation, storytelling, and fairness in artificial intelligence.
Psychological Brain & Sciences,& Anthropology: Caroline Robertson, Emily Finn, & Zaneta Thayer
Computational Mapping of Multisystem Brain Dynamics Across Pregnancy and Postpartum
Pregnancy is one of the most profound biological transitions in human life, involving coordinated changes in hormones, vascular physiology, sleep, cognition, and brain organization. Despite its ubiquity and importance, we still lack a systems-level computational account of how these processes reorganize during pregnancy, how they recover postpartum, and how these biological dynamics give rise to changes in cognition. This project addresses that gap by following the same individuals intensively across the transition into and out of pregnancy,
beginning before conception and continuing through postpartum recovery. Across approximately 24 precision MRI measurement sessions, the team is collecting measurements of functional brain dynamics, vascular dynamics, sleep physiology, endocrine rhythms, and cognition. By developing computational models that integrate these data streams, the project aims to reveal how brain networks, bodily physiology, endocrine rhythms, sleep, and cognition change together over time; how these systems return toward an individual's pre-pregnancy
baseline; and which changes are temporary versus persistent. Together, this work will establish a computational framework for understanding pregnancy as a coordinated multi-system transformation that reshapes brain, physiology, and cognition across an extended postpartum arc.
Psychological Brain and Sciences: Viola Stoermer & Yong Min Choi
Computational Principles of Multisensory Spatial Attention in Human Brain
How the brain selects relevant information from a continuous stream of multisensory input remains a central open question in cognitive neuroscience. While most computational models of attention have focused on vision in isolation, the mechanisms underlying cross-modal attention are still poorly understood. This project addresses this gap by testing a model of spatial attention that operates jointly across vision and audition.
The research combines population receptive field modeling of fMRI data with large-scale functional connectivity analyses to examine whether auditory spatial attention recruits the fine-grained spatial architecture of the visual cortex—and how this interaction reshapes whole-brain network organization.
Led in part by postdoctoral fellow Yong Min Choi, who serves as co-PI, the project brings additional expertise in computational neuroimaging and systems neuroscience to the effort.
The resulting framework will provide a mechanistic, computational account of how the brain integrates information across sensory modalities to guide perception and behavior.
Psychological Brain and Sciences: Mark Thornton
Stress-testing social interactomics in the field: Dialect creation in Un Nuevo Amanacer
In the Santiago, Chile neighborhood of Un Nuevo Amanecer, interactions among diverse immigrant communities are giving rise to an emergent dialect of Spanish. This project applies a novel computational framework—social interactomics—to investigate how such dialects form and evolve.
Using a suite of deep learning tools, the research will quantify key dimensions of social behavior, including facial expressions, vocal tone and prosody, body pose and gesture, and the semantic content of speech. By integrating these multimodal signals, the project seeks to explain linguistic change as a function of the social cues individuals produce and encounter in their everyday environments.
This approach offers a new, data-driven account of dialect formation, linking language evolution directly to patterns of human interaction.
Film & Media Studies: Mark Williams
The Media Ecology Project (MEP)
The Media Ecology Project (MEP) is Dartmouth-originated pioneering digital humanities initiative for the study of historical moving images, though the full potential of its archival collections has remained constrained by the limits of traditional cataloging. This project will address that gap by developing AI-powered tools that automatically generate rich, time-based metadata — identifying objects, settings, and situations within films — making vast and previously inaccessible archival collections discoverable by scholars worldwide.
The research will draw on MEP's collection of roughly 3,000 public-domain U.S. Information Agency documentary films as a development corpus, combining computer vision methods with Dartmouth's existing research infrastructure.
The final output will provide a scalable model for archival search and discovery that can be extended to film collections globally. In doing so, the project will offer scholars, educators, and archivists across MEP's growing international network — spanning Europe, Africa, East Asia, and Latin America — powerful new tools for engaging with moving image history.