Developed an Embedding Statistics Tracking Framework for the pytorch/FBGEMM repository, focusing on enhancing observability and performance analysis of embedding workloads. The work involved implementing a dedicated stat library in C++ with CMake, enabling the collection and logging of embedding access patterns. Instrumentation was designed to be controlled via environment variables, allowing performance metrics to be gathered without modifying application code. This approach established a foundation for future data-driven optimizations in embedded systems and system programming contexts. The solution aligned with existing instrumentation standards, supporting maintainability and extensibility while addressing the need for actionable insights into embedding workload performance.