During March 2026, João Vabo focused on backend robustness for the cvxgrp/cvxpy-ipopt repository, addressing a critical bug in the COO tensor backend. He fixed an issue in the coo_mul_elem function, ensuring correct broadcasting of dimensions when multiplying COO tensors with varying shapes. Using Python and numpy, João implemented both unit and integration tests to validate shape inference and broadcasting semantics, aligning output shapes with SciPy-style broadcasting. His work reduced shape-related errors in optimization workflows and improved the reliability of tensor operations. The changes were thoroughly documented and conformed to repository standards, laying groundwork for future backend enhancements.