Worked on the pytorch/pytorch repository to deliver Inductor backend extensibility through the implementation of custom FX pass registration and graph caching. Developed the CustomGraphModulePass interface, enabling backend-specific FX passes and efficient caching of compiled graphs to reduce recomputation and support backend experimentation. This work focused on strengthening the backend architecture, allowing for faster iteration cycles and future optimizations. Leveraged Python and PyTorch FX for graph processing and optimization, with an emphasis on backend development and testing. No major bugs were addressed during this period, as the primary contribution centered on enhancing extensibility and performance potential for Inductor backends.