They asked me to implement a sparse matrix from scratch, without relying on any existing matrix or linear algebra libraries. This required designing an efficient internal data structure to store only the non-zero elements, rather than allocating memory for the entire matrix. In addition to the core representation, I needed to implement both addition and multiplication operations, making sure they handled sparsity correctly, maintained good performance, and produced accurate results even when matrices had different sparsity patterns.