Prediction alone is not enough
In healthcare, accuracy is not the only measure of quality. A clinical prediction should be interpretable, placed in context, and accompanied by its uncertainties and limitations.
Useful healthcare AI must understand time, explain its outputs, and leave clinical decisions to professionals.
My work starts from a simple premise: a clinical prediction has value only when it is grounded in the patient’s trajectory, its rationale can be understood, and its implications can be discussed by those delivering care.
Predicting an event is not the end goal.
The real challenge is understanding what is changing, why it matters, and how that information can support a clinical decision.
Three principles
These principles guide my research, technical choices, and approach to applied R&D projects.
In healthcare, accuracy is not the only measure of quality. A clinical prediction should be interpretable, placed in context, and accompanied by its uncertainties and limitations.
Care unfolds over time. Symptoms, treatments, responses, and decisions form a trajectory that models must represent beyond isolated snapshots.
A decision-support system should inform, not dictate. It should expose the evidence behind an output, make its limits visible, and preserve the professional’s ability to question or override it.
These principles come from more than academic research.
Nearly 20 years in digital technologies taught me that a system is valuable only when it answers a real need and can be understood by the people who use it. Teaching reinforced the discipline of explaining complex ideas clearly; doctoral research and a University Diploma in Critical Care brought those principles into direct contact with clinical realities.
This blend of scientific, technical, and clinical perspectives shapes how I work: understand before predicting, explain before automating, and design with the people who will actually use the system.
I am not only interested in whether a model predicts correctly.
I want to understand what it has learned from the patient’s trajectory, why it anticipates a change, and how to make that information useful without replacing clinical judgment.
These principles guide both my research and my applied R&D work on patient trajectories, temporal models, and explainable clinical AI.