Worked on the llvm/torch-mlir repository to enhance compatibility between Torch index-like operations and StableHLO, focusing on improving the lowering path to StableHLO’s gather and scatter operations. The work involved refining broadcasting semantics and shape handling so that tensors align correctly during these operations, reducing dimension-mismatch issues and deployment friction for Torch models targeting StableHLO-backed backends. Leveraged C++, MLIR, and advanced tensor manipulation techniques to implement robust broadcasting and dimension alignment, ensuring smoother interoperability. This feature addressed a key integration challenge, enabling more reliable deployment of index-like operations from Torch to StableHLO while maintaining correctness and stability in model execution.