Research

My research develops AI and data science methods to extract, fuse, and clinically validate digital signals from everyday life and clinical assessments. The goal is to build models of neurological disease that are rigorous enough to translate into clinical practice.

Multimodal Digital Phenotyping

The central question driving my work is how to characterise neurological disease from the combination of passive and active digital measurements. This means integrating smartphone-based clinical tests, wearable accelerometry, neuromotor digital assessments, standardised questionnaires, and clinical records. The longer-term ambition is toward multimodal integration with neuroimaging and biofluid biomarkers.

My doctoral work (Orange Labs / Université Grenoble-Alpes, 2017–2020) established that passive mobile phone data encodes meaningful behavioural biomarkers. Circadian rhythm signatures and social asymmetry patterns were shown to be associated with depression and behavioural decline in elderly populations (8 papers; 1 patent). My postdoctoral work extended this framework to wearable-based gait phenotyping. Drawing on both my research and industry experience, my current work at Oxford is organised around two main axes:

Motor Trajectory Stratification and Disease Progression

Neurological diseases are deeply heterogeneous. Understanding how individual patients progress over time is one of the harder problems in clinical research. A focus of my work is developing tools to stratify motor trajectories in Parkinson’s disease using unsupervised machine learning. This characterises how patients cluster by the shape and speed of their progression. I then combine these approaches with smartphone-based clinical assessments to evaluate their ability to predict future motor progression from a single baseline visit. This is the focus of my current paper in BMJ Neurology Open (2026, under review).

Composite Scores and Clinical Validation of Digital Tools

The Oxford Parkinson’s Disease Centre has a strong tradition in developing composite scores for Parkinson’s disease. These offer a principled way to fuse data from multiple domains, guided by domain expertise and clinical constraints, an approach well established in the literature. I extend this work to new Digital Health Technologies, including smartphone-based clinical assessments. The goal is to develop validated clinical scores that match current standards but can also be used outside research settings, in real-life conditions. This motivates a 20+ site collaboration toward the largest multimodal clinical and digital dataset in Parkinson’s disease research to date. Early evidence of its value is demonstrated on a subcohort in our work on smartphone-based dopaminergic prediction.

Agentic AI and the Next Step for Digital Health

The rise of foundation models and agentic AI represents a significant shift for this field. Harnessing these approaches alongside existing analytical pipelines opens new opportunities for multimodal data analysis in neurology. It enables more systematic and autonomous exploration of large heterogeneous datasets, going beyond what hypothesis-driven analyses alone allow. This is a space I try to stay at the frontier of, both in practice and in how I communicate it to the clinical community. My recent finalist position at the Oxford Clinical AI Hackathon reflects that ongoing effort.

Inclusive Research and Global Collaboration

The global collaboration network underlying this research is central to the scientific ambition. Despite the promise of new methods and technologies, the critical resource in this field remains data: diverse, longitudinal, well-characterised data from real patient populations. Building that requires broad collaboration across clinical and academic institutions.

There is also a dimension I find genuinely meaningful. Everyday digital tools offer an opportunity to extend monitoring and assessment to populations with limited access to specialist care. Innovation in digital health can carry real social value, for patients, for healthcare systems, and for communities where the burden of neurological disease is high but the resources to address it are scarce.