Worked on the pymc-devs/pytensor repository to enhance the reliability of gradient checking workflows in probabilistic modeling. Addressed a usability issue in the verify_grad function by introducing a default random number generator when none is provided, eliminating the need for users to supply a RandomState instance and preventing related TypeErrors. This change improved the stability and ergonomics of the API, making it easier for downstream projects to integrate PyTensor with PyMC. The work focused on Python and numerical computing, applying skills in gradient checking to strengthen forward-compatibility and reduce friction for users verifying gradients in complex modeling pipelines.