We present a real-time structural elastic model of the Research Vessel (RV) Gunnerus. The RV Gunnerus is a research vessel owned and operated by the Norwegian University of Science and Technology in Trondheim, Norway. In addition to serving vital ocean research roles, owing to the many onboard sensor data streams which are freely accessible, the vessel also serves as a suitable physical asset for high-TRL digital twin applications. Our aim is the implementation of a digital twin for monitoring the structural integrity of the vessel, aimed at predictive maintenance for ships. To that end, we demonstrate an efficient lightly intrusive reduced order model (ROM) with AI-enhanced capabilities for operator inference and accuracy correction using the corrective-source term approach (CoSTA). The lightly intrusive concept for ROMs establishes a common data interchange framework for high-fidelity models and ROM implementations, and defines a minimal set of exported data from a high-fidelity model that must be met to allow construction of non-intrusive ROMs. The integration of hybrid AI-driven methodologies will be shown to not only enhance the lightly intrusive ROM concept, but also to be crucial for the high performance and accuracy of the model, which would not have been possible using conventional ROM methods, and which enables the higher strata of digital twin functionalities.