This project applies unsupervised machine learning to investigate how students interact with ChatGPT during educational tasks. By clustering user behaviors, prompt structures, and interaction dynamics, the system will surface archetypes of AI reliance — such as synthesis avoidance, prompt mimicry, or overconfidence in outputs. These latent usage structures will be mapped to pedagogical pathologies, revealing how LLMs may unintentionally reinforce shallow learning. The ultimate goal is to generate guidelines, interventions, and feedback systems that promote reflective and generative AI use in education.

OpenStetho is a digital stethoscope prototype with onboard audio capture, amplification, and embedded signal processing. The device integrates a microcontroller for real-time filtering or feature extraction and streams data wirelessly or via wired interface to a companion application. Designed as an open-source biomedical hardware platform, OpenStetho enables advanced features such as sound separation, playback, and diagnostic augmentation, supporting both educational and clinical applications.

Preekit is a low-cost, passive diagnostic tool embedded in a sanitary liner that detects elevated protein levels in urine — an early warning sign of preeclampsia. By enabling discreet, color-change–based monitoring during daily wear, Preekit reduces the need for urine handling or frequent clinic visits. Its intuitive design supports maternal self-monitoring and provides earlier opportunities for intervention, particularly in underserved or rural communities where prenatal care access is limited.

This project develops novel error-correcting code strategies specifically designed for short DNA sequences used in visual encoding applications. By modeling DNA encoding as a noisy communication channel subject to visual degradation, printing errors, and biological mutations, the research aims to create robust coding schemes that can withstand realistic multi-modal noise while respecting biological constraints such as GC content balance and homopolymer limitations. The work addresses critical gaps in DNA data storage research, particularly for applications involving short sequences and mixed error types.

SafeSip Tag is a passive water contamination monitoring device that clips onto or floats in household water containers. The device continuously monitors water quality through colorimetric indicators that respond to bacterial byproducts and chemical contaminants, providing real-time visual feedback without user intervention. This low-cost solution addresses the critical need for accessible water quality monitoring in communities with limited access to laboratory testing.

TempTrace is a low-cost, passive wearable strip with thermochromic zones that activate sequentially over time, creating a visible timeline of body temperature. By distinguishing between transient spikes and sustained fever, TempTrace enables caregivers to better triage illnesses such as malaria or sepsis. Designed for at-home use in low-resource settings, it removes the need for thermometers or digital tools, offering a simple, intuitive solution for fever monitoring in children and vulnerable populations.

TremorEase is an assistive AI system that stabilizes stylus and touchscreen input for individuals with tremors (e.g., Parkinson’s disease). The system captures stylus movement in real time, detects involuntary tremor-induced noise, and corrects it using adaptive ML and signal filtering. Unlike diagnostic or hardware-only solutions, TremorEase is designed for everyday use, ensuring smoother, more accurate writing, drawing, and interaction across modern touchscreen devices. With customizable filters, cross-platform support, and real-time feedback, it provides a highly flexible accessibility solution for productivity, communication, and creative expression.

VeriTalk is a real-time conversational assistant that passively listens to dialogue, detects factual claims, and retrieves credible sources or corrections on the fly. Unlike traditional after-the-fact fact-checking, VeriTalk provides in-conversation epistemic support — surfacing statistics, definitions, or corrections without disrupting conversational flow. The system empowers users in meetings, classrooms, or everyday discussions by acting as a “quiet, continuous second brain” that anchors reasoning in evidence and provides post-conversation breakdowns of claims with confidence scores and citations.