Worked on the ROCm/aiter repository to deliver a scalable intranode All2All communication manager for Mixture-of-Experts workloads, leveraging FlyDSL and custom MLIR-based kernels to replace existing Mori primitives while maintaining shared memory heap compatibility. Implemented JIT-compiled dispatch and combine kernels supporting multiple data types and quantization modes, enhancing flexibility for distributed systems. Additionally, stabilized the CI smoke workflow by updating ATOM main branch tracking and constraining matrix generation to a single 1p1d configuration, which reduced nightly failures on unsupported multi-node topologies. The work demonstrated depth in CI/CD, kernel development, and Python, focusing on robust infrastructure and maintainable workflows.