Worked on the pytorch/torchrec repository to enhance memory profiling and metric computation performance for machine learning workloads. Focused on improving the accuracy of memory usage metrics by moving memory checks inside the trace execution path, which allowed for more reliable detection of memory regressions during compute operations. Addressed a bug that previously caused skewed memory measurements, aligning metric collection with actual execution. Later, optimized the RecMetricModule by removing unnecessary memory checks, reducing overhead and accelerating metric computation. All changes were implemented in Python, leveraging skills in performance optimization and unit testing, and were tracked through clear, reviewable Git commits.