Worked on the espressif/llvm-project repository to enhance the MLIR GPU backend by adding support for bf16 and i1 data types in the gpu::shuffle operation during conversion to LLVM SPIR-V. The implementation involved introducing bitcasting and type extension or truncation logic, converting bf16 to i16 and i1 to i8 before invoking SPIR-V intrinsics, then restoring the original types afterward. This approach improved correctness and expanded the range of supported workloads for SPIR-V targets, reducing manual type handling for downstream users. The work demonstrated skills in compiler development, GPU programming, and C++ within the context of MLIR and SPIR-V technologies.