Worked on the google/heir repository to deliver a tensor reduction optimization aimed at improving the efficiency of tensor-heavy workloads. The approach involved implementing a linalg.reduce transformation in C++ to convert eligible linalg.generic operations, enabling a rotate-and-reduce sum mechanism for more effective tensor reductions. Additionally, introduced the ReductionCanonicalizations pass to ensure correctness and facilitate further optimization within the MLIR pipeline. This work enhanced both performance and maintainability of the optimization process, leveraging expertise in C++ development, MLIR, and tensor operations. No major bugs were reported during this period, reflecting a focused and well-executed engineering effort.