We introduce a GENERIC informed neural network (GINN) closure methodology for viscoelastic flows that learns thermodynamic structure preserving constitutive relations that can be can be embedded inside the RheoTool/OpenFOAM finite volume solver. We extend the methodology proposed in [1], to enable learning of arbitrary tensorial constitutive models incuding both, purely entropic and dissipative tensorial terms from data. The methodology proceeds in two stages: in Stage 1, the polymeric entropy is learned from stress data, using automatic differentiation to recover the entropy conjugate that provides the corresponding GENERIC thermodynamic consistency [1]; in Stage 2, the mobility friction matrix is inferred from the evolution of the conformation tensor through residual minimization [2]. Both neural blocks are evaluated cell by cell within RheoTool without modifying the finite-volume discretization. The data driven constitutive equation approximation is validated on the flow around a cylinder benchmark, where the two stage closure reproduces the reference finite volume solutions with accurate fields over a range of Weissenberg numbers. The learned closure’s transferability to new cases is demonstrated by numerically solving a cross slot configuration. Across these tests, the GINN closure closely matches the baseline model and remains numerically stable, indicating that the thermodynamically-constrained enables geometry-agnostic viscoelastic constitutive relations reusable across flow rates and geometries without the need of retraining. REFERENCES [1] DN Simavilla, A Bonfanti, I García-Beristain, P Español, M Ellero. Hammering at the entropy: a GENERIC-guided approach to learning polymeric rheological constitutive equations using PINNs Journal of Fluid Mechanics p.1016, A11, 2025. doi.org/10.1017/jfm.2025.10325 [2] I García-Beristain, DN Simavilla, M Ellero, A GENERIC informed neural network methodology to learn thermodynamic structure-preserving constitutive relations, in preparation (2026).