Andrija Bosnjakovic developed a core feature for the llvm/torch-mlir repository, focusing on the integration and lowering of stochastic ReLU variants within the Torch MLIR dialect. Using C++ and Python, Andrija implemented support for torch.aten.rrelu_with_noise and its backward operation, enabling more robust and optimized machine learning workflows. The work included refining the TorchToLinalg lowering path, which improved correctness and reliability for downstream pipelines. By enhancing MLIR-based tooling readiness, Andrija’s contribution addressed integration challenges for models using stochastic ReLU, supporting easier adoption and performance optimization. The depth of the work demonstrated strong proficiency in MLIR and machine learning systems.