Worked on backend enhancements for the llvm/torch-mlir repository, focusing on improving model interoperability and execution reliability between PyTorch and StableHLO. Developed and tested new operator support, including lowering for aten.view.dtype, adaptive pooling, and tensor shape inference, using C++ and MLIR. Expanded the operator set with implementations for loss functions and tensor operations such as isfinite, column_stack, float_power, and threshold, ensuring robust input validation and comprehensive test coverage. Emphasized stability and correctness by strengthening validation and conversion patterns, enabling more seamless end-to-end model deployment workflows. Demonstrated depth in C++ development, PyTorch integration, and tensor manipulation.