Contributed a targeted backend enhancement to the ignaciosica/tinygrad repository by implementing support for the NONZERO operation in the ONNX backend. This feature enables extraction of non-zero tensor indices, expanding the framework’s tensor manipulation capabilities and improving ONNX model compatibility. The work was delivered in Python and focused on backend development and machine learning frameworks, with careful attention to maintainability and future extensibility. By broadening model interoperability and laying the groundwork for additional indexing operations, the contribution addressed a specific gap in ONNX support. All changes were consolidated in a single, traceable update to facilitate code review and ongoing maintenance.