Worked on performance profiling and control systems across two major open-source repositories. In Lightning-AI/pytorch-lightning, developed checkpoint saving performance profiling by instrumenting the Trainer.save_checkpoint method in C++ and Python, enabling measurement of checkpoint overhead and identification of bottlenecks during large-scale model training. This provided actionable insights for optimizing training throughput and reducing latency. In RobotLocomotion/drake, standardized torque sign conventions across IIWA, Jaco, and Panda robots, updating middleware and documentation to ensure consistent torque reporting. Enhanced developer onboarding by improving documentation and traceability. Demonstrated expertise in control systems, middleware, and model checkpointing, with a focus on maintainable, measurable solutions.