Aman Singh developed an observability feature for distributed training workloads in the aws-samples/awsome-distributed-training repository. He implemented NCCL metrics collection using Python and Bash, integrating NCCL Inspector with Prometheus via node_exporter's textfile collector. His approach included updates to the observability stack and Slurm integration, ensuring metrics were gathered without affecting non-metrics runs. Aman extended infrastructure scripts and workflows to support metrics collection on compute nodes, adding gating and configurable intervals for data dumps. This work provided end-to-end visibility into NCCL communication, laying a foundation for data-driven performance optimization and faster diagnostics in large-scale distributed training environments.