Clinical & precision-health AI

Clinical AI is not judged on a held-out test set; it is judged on whether a surgeon leaves the feature switched on during a live procedure. We build machine-learning systems for the point of care — where latency is a hard constraint, hardware is fixed, cloud is off the table, and a single unreliable frame erodes trust that took months to earn.

The work spans real-time image segmentation for surgical guidance, multi-biomarker risk scoring that turns a routine blood panel into a multi-dimensional health profile, and voice biomarkers for digital health. In every case the modelling is the smaller half of the problem: data governance, edge optimisation, and reliability engineering are what turn an accurate model into a deployed clinical tool.