Developed and integrated NCCL metrics collection and exposure for Prometheus within the aws-samples/awsome-distributed-training repository, focusing on enhancing observability for distributed training workloads. Leveraged Python and Bash to extend the monitoring stack, enabling end-to-end visibility of NCCL communication across compute nodes using NCCL Inspector and node_exporter’s textfile collector. The implementation included updates to lifecycle scripts and install workflows, with configurable observability flags and metrics dump intervals to ensure minimal impact on non-metrics runs. This work established a foundation for data-driven performance optimization and streamlined diagnostics in large-scale training environments, emphasizing DevOps practices and robust scripting for infrastructure automation.