Vesselity Maritime Analytics builds HullSight, a platform for assessing the condition of ship hulls. A hull is a fuel question: fouling increases drag, and cleaning too late means months of burning more than necessary. The answer used to be dry-docking, at over 50,000 dollars per inspection and with the vessel earning nothing while it sits.

We have been working on the platform behind that idea since early 2026. A segmentation pipeline in Python turns underwater footage into an assessment of which areas of the hull are fouled and how badly. We helped build the fouling detection models, including synthetic training data for the cases where real imagery was too sparse, and validated the results against the judgement of domain experts.

The frontend is a React monorepo that has grown across several versions. What mattered was less the model accuracy than how a finding is presented so an operator can act on it: where exactly the fouling sits, how it developed since the last survey, and whether that justifies a cleaning.