A computational fluid dynamics-based framework for optimizing 3D geometry in an idealized total cavopulmonary connection (TCPC) is presented. The TCPC is a surgical procedure designed to treat congenital heart defects involving a single functional ventricle. The presented custom optimization framework integrates Python-based geometry generation, lattice Boltzmann method (LBM) simulations, and gradient-free optimization algorithms, including Nelder-Mead and the Mesh Adaptive Direct Search methods. The three optimization steps – generation of parameterized 3D geometry, simulation of incompressible Newtonian fluid flow with a rigid wall, and evaluation of objective functions – are executed automatically. The massively parallel implementation of LBM on GPUs allows the use of a spatial resolution suitable for optimizing the flow metrics sensitive to the actual resolution, such as the turbulent kinetic energy or near-wall shear rate. A simplified, parameterized model of the TCPC geometry was used to test the framework, demonstrating its feasibility and effectiveness. While this study focuses on idealized geometries with simplified assumptions, the results provide a foundation for extending the framework to patient-specific data and more complex physiological scenarios. This work represents a step in applying computational optimization to cardiovascular surgery, with the potential to improve clinical outcomes and patient-specific treatment planning.