Worked on the intel/mlir-extensions repository to enhance GPU code generation and vectorization workflows. Developed a new IndexCastOp conversion pattern within the XeGPUToVC pass, improving index casting in GPU vector computations and clarifying conversion semantics for maintainable compiler pipelines. Addressed reliability by fixing the vector linearization process to support f16 LoadOp operations, expanding hardware compatibility and ensuring correctness for half-precision floating-point workloads. Employed C++, MLIR, and compiler design expertise to deliver targeted features and bug fixes, with a focus on robust, test-driven development. The work demonstrated depth in GPU programming and vectorization, directly supporting MLIR extension goals.