Computational and experimental modeling in cardiovascular fluid mechanics has provided valuable fluid mechanics-based biomarkers that can be used in evaluating cardiovascular disease severity and treatment planning. Given the limitations of these experimental and computational models, there is growing interest in using machine learning to address these limitations. In this talk, I will summarize some of our group's recent work in scientific machine learning and their applications in blood flow modeling. I will focus on different data regimes ranging from large data to no data. I will discuss different appropriate machine learning approaches and the associated challenges. Specifically, I will present examples related to data- driven reduced-order modeling (ROM), deep learning, physics- informed machine learning, and differentiable programming.