Developed end-to-end performance profiling enhancements for distributed reinforcement learning workloads in the volcengine/verl repository. Focused on integrating NSYS and NSight-based profiling tools within Python-based environment workers, the work enabled detailed, cross-worker performance analysis and runtime configurability. Implemented coordinated profiling across all worker groups, adding decorators and extensions to key workflow stages for improved observability. Enhanced the controller and trainer modules to support timeline profiling outputs and runtime environment options, allowing targeted diagnostics throughout RL training cycles. These changes improved visibility into distributed system performance, facilitating faster root-cause analysis, reduced training time, and more efficient resource utilization in machine learning workflows.