PINA is a new open-source Python library that enhances Scientific Machine Learning (SciML) by streamlining the process of modeling and simulating complex scientific systems. Built on PyTorch and PyTorch Lightning, it integrates seamlessly with PyTorch Geometric, allowing researchers to tackle a wide variety of scientific challenges, from solving partial differential equations to simulating dynamics and modeling object deformations. PINA organizes the SciML workflow into four main stages: problem definition, model design, solver selection, and training, each designed to maintain structure without sacrificing flexibility. The problem definition phase allows users to formalize their system's mathematical representation, accommodating various data types, including tensors and graph-structured inputs. In model design, users can choose from built-in architectures or create custom neural network models to map input variables to outputs. The solver selection connects the defined problem to the model's learnable parameters, offering a range of strategies like Physics-Informed Neural Networks and supervised solvers. Training leverages PyTorch Lightning's capabilities for efficient model execution across different devices. PINA's modular design supports reproducibility and extensibility, making it suitable for both academic research and industrial applications. It can handle a broad spectrum of tasks, including physics-based modeling and equation discovery. Users can get started with easy installation and access tutorials and documentation for guidance. Community contributions are encouraged, allowing users to report issues, add features, or assist with maintenance.