Worked on the pytorch/pytorch repository to address a critical C++ code generation issue affecting PyTorch Inductor. Focused on resolving a parsing bug where constructor-call expressions were misinterpreted as function declarations, which caused compile-time failures in array and tensor output handling. The solution involved updating code generation logic to use brace initialization, thereby eliminating the Most Vexing Parse and improving stability for AOTI lowering. Added targeted unit tests in Python and C++ to validate the new initialization path and prevent regression. Changes were integrated into internal pipelines and validated through automated test infrastructure, ensuring robust and maintainable code improvements.