Physics-Informed Neural Network (PINN) has emerged as a powerful simulation tool in numerous scientific, industrial, and engineering applications for performing forward and inverse modeling. However, applications of PINNs to cardiovascular blood flow modeling have been limited due to the lack of dedicated simulation platforms. We have developed an open-source, validated, and modular platform, CardioPINNs, that is highly efficient in solving inverse problems by integrating clinically acquired sparse data (e.g., imaging and clinical parameters) with physics-based simulations. In particular, we have implemented several features in CardioPINNs for robust vascular blood flow simulations, including pulsatile inflow boundary conditions; reduced-order models for outlet boundary conditions; integral boundary losses to further regularize training; a preprocessing framework to generate point clouds and surface and volumetric meshes for visualization; and a post-processing framework to compute hemodynamic biomarkers such as wall shear stresses, oscillatory shear index, kinetic energy, turbulent dissipation, and others. We have verified CardioPINNs under a range of flow conditions (from laminar to quasi-turbulent) and evaluated its performance across several anatomical geometries, including cerebral aneurysms, abdominal aortic aneurysms, and left ventricles. Lastly, we validated CardioPINNs against high-resolution in vitro 4D Flow MRI velocity obtained from an aortic phantom. To highlight the inverse modeling capabilities of CardioPINNs, we demonstrate its ability to reconstruct complete three-dimensional velocity fields using sparse centerline measurements away from the vessel wall, as well as to accurately predict near-wall hemodynamics. The CardioPINNs platform provides a comprehensive and validated pipeline for conducting cardiovascular physics-informed machine learning simulations at the intersection of scientific computing and machine learning.