During the month, contributed to the microsoft/mscclpp repository by addressing a critical kernel-level synchronization issue in the allreduce8 operation. Focused on improving concurrency reliability and data integrity in high-performance distributed training workloads, the work involved implementing precise thread synchronization to ensure all memory writes completed before signaling dependent threads. This approach eliminated race conditions and preserved correct data ordering, directly enhancing the correctness of parallel computations. The solution was developed using C++ and CUDA, leveraging expertise in GPU programming and parallel computing. The fix was validated through targeted testing and code review, maintaining performance goals while reducing nondeterminism in distributed workflows.