In February 2026, Sanspras Prasobh enhanced the PyTorch Profiler within the pytorch/pytorch repository by integrating compilation events into its profiling output. Using Python and leveraging backend development and performance profiling skills, Sanspras introduced a thread-local logging mechanism that captures detailed compilation phases during profiler traces. This approach improved profiling accuracy by isolating events to individual threads, reducing cross-thread noise and enabling developers to analyze compilation performance more effectively. The work addressed a key need for better observability in Dynamo-backed workflows, providing tangible value by accelerating optimization cycles and supporting more precise root-cause analysis during performance debugging and testing.