Worked on enhancing the NVIDIA/NeMo-Run repository to improve reliability and scalability for high-performance computing environments. Focused on distributed systems challenges, the work addressed multi-node execution by refining preparation orchestration so that initialization steps occur only on the primary process, reducing redundant operations during large-scale experiments. Leveraged Python to implement environment-aware node ranking, specifically integrating SLURM compatibility by detecting SLURM_NODEID for accurate node assignment. These changes stabilized multi-node launches and improved throughput in cluster settings. The engineering approach emphasized robust system administration practices, ensuring that experiments run more predictably and efficiently across diverse distributed computing infrastructures without unnecessary resource waste.