Worked on performance optimization and cloud-native tooling, contributing to both ROCm/onnxruntime and conda-forge/staged-recipes repositories. Developed a memory-efficient method in C++ for ROCm/onnxruntime, enabling direct reading of tensor attributes from raw data to reduce memory usage during ONNX model session creation. In conda-forge/staged-recipes, created and maintained a Kyverno CLI recipe, establishing build instructions, dependency management, and license compliance through explicit pinning and governance updates. Leveraged skills in C++, Go, and DevOps practices to improve packaging quality, reproducibility, and audit readiness. Focused on scalable deployment, open source contribution, and robust maintenance for machine learning and policy management tools.