Worked on enhancing backend stability and compatibility for the Triton AMDGPU/LLVM backend, focusing on maintaining correctness as LLVM evolved. Addressed a bug in the AMDGPU v_permlane_swap path within the triton-lang/triton repository, improving the reliability of backend code generation. Updated the LLVM integration by bumping the project to a newer commit, ensuring alignment with upstream changes and forward compatibility. Introduced API-level improvements, such as adding an IRBuilder header and a default parameter to convertFuncOpToLLVMFuncOp, which streamlined backend integration across LLVM and MLIR bridges. Utilized C++ and leveraged expertise in backend and compiler development throughout the process.