Background¶
I am a computational neuroscientist and machine learning researcher working at the intersection of artificial intelligence, clinical neuroscience, and multimodal data integration. My research focuses on developing explainable machine learning frameworks that clinicians can genuinely trust and act on — not just models that perform well on benchmarks.
I hold a Ph.D. in Hearing Science from Western University, where my dissertation focused on speech intelligibility assessment using automatic speech recognition, spanning deep learning for audio processing, acoustic modeling, and psychoacoustics. Prior to that, I completed an M.Sc. in Mechatronics Engineering with a focus on signal processing, and a B.Sc. in Computer Engineering with a specialization in embedded systems.
My technical experience spans transformer-based speech models, multimodal sensor fusion, on-device machine learning for iOS and embedded systems, and end-to-end ML pipeline development using TensorFlow, PyTorch, and scikit-learn. I have also co-founded Sensalog Inc., a digital health company developing a wearable medical device platform for at-home monitoring of Parkinson’s disease symptoms, where I served as CTO and led the design and deployment of real-time multimodal ML systems — an experience that deeply shaped my appreciation for the gap between research and clinical deployment.
I am currently a CNS Cohort 2 Postdoctoral Fellow at Western University, working under the supervision of Dr. Angela Roberts in the School of Communication Sciences and Disorders. My fellowship is funded by The Hilary & Galen Weston Foundation (2026–2028).
Project¶
Neurodegenerative diseases such as Alzheimer’s disease, Parkinson’s disease, and vascular cognitive impairment represent a growing challenge for aging populations worldwide. Each condition affects the brain differently — disrupting speech, movement, cognition, and brain structure in distinct but often overlapping ways — making early and accurate diagnosis a significant clinical challenge.
This project develops a multimodal machine learning framework for the early detection, differential diagnosis, and monitoring of disease progression across these three conditions. Rather than relying on a single data source, the framework integrates four complementary modalities: speech and acoustic features, motor and gait signals, cognitive assessment data, and neuroimaging biomarkers. By fusing these data streams into a unified model, the goal is to capture a richer and more complete picture of each patient’s neurological profile than any single modality could provide alone.
A central design principle of this framework is clinical interpretability. The framework is built to reveal not only what prediction a model makes, but which features — and which combinations of features across modalities — are driving that prediction. This level of transparency is essential for clinical trust and adoption: a model that cannot explain its reasoning will not be used, regardless of its accuracy.
The framework is developed using two major Canadian open neuroscience datasets — the Ontario Neurodegenerative Disease Research Initiative (ONDRI), accessed via Brain-CODE, and the Comprehensive Assessment of Neurodegeneration and Dementia study (COMPASS-ND), accessed via LORIS. Together, these datasets provide rich, longitudinal, multimodal data across the target diagnostic groups and healthy controls. A key validation goal is cross-dataset generalizability — ensuring that models trained on one dataset transfer effectively to the other, a critical requirement for real-world clinical deployment.
Ultimately, this work aims to produce tools that support clinicians in identifying neurodegenerative disease earlier, distinguishing between conditions that present similarly, and tracking how those conditions evolve over time — with the transparency needed to make those tools genuinely trustworthy in practice.