Worked on the root-project/root repository to enhance the reliability of ONNX graph transformations by addressing a subtle bug in the fusion process. Focused on C++ development and graph parsing, the work involved correcting node pairing when ONNX graphs are reordered, which previously led to mispairing due to mismatched child list indexing. By aligning child lists with the parsed node order using nodesOrder[i], the fix ensured that nodesChildren corresponded accurately to the current graph structure. This improvement reduced fusion errors in production environments and strengthened the stability of machine learning model execution paths that depend on fused ONNX graphs.