Worked on the deepinv/deepinv repository to expand TomographyWithAstra, enabling support for non-cubical 3D phantoms by extending input and output shape handling. This enhancement allows researchers to model a broader range of tomography scenarios with greater realism. The implementation involved modifying Python code to accommodate flexible data shapes, adding unit and integration tests to ensure robust handling of non-cubic signals, and updating both the changelog and inline documentation for clarity. Collaborated with other contributors through code reviews and maintained adherence to project quality standards, leveraging skills in Python, data science, and machine learning throughout the development process.