Worked on the llvm/torch-mlir repository to enhance the TorchToTosa conversion path, focusing on both feature development and reliability improvements. Implemented 1D convolution support by normalizing 1D inputs to 4D tensors, leveraging existing conv2d lowering logic in C++ and MLIR to extend model compatibility. Addressed cross-platform test stability by refining test infrastructure and adopting CHECK-DAG for deterministic validation across Linux, macOS, and Windows. Fixed critical lowering issues, such as dtype and rank mismatches in AtenWhereSelfOp, and restored end-to-end testing coverage for key modules. Utilized Python scripting and software testing to ensure robust, maintainable deployment workflows.