Mostafa contributed to the intel/mlir-extensions repository by enhancing GPU code generation and vectorization workflows. He developed a new IndexCastOp conversion pattern within the XeGPUToVC pass, improving the handling of index casting in GPU vector computations and clarifying conversion semantics for more robust compiler pipelines. Using C++ and MLIR, he also addressed a critical bug in the vector linearization process, enabling support for half-precision (f16) LoadOp operations and adding targeted tests to validate correctness. His work deepened the reliability and maintainability of vectorized floating-point handling, aligning with broader hardware compatibility and performance goals in compiler design.