Worked on the ROCm/madengine repository to enhance stability and observability by unifying performance log parsing across multiple environments. Addressed a complex bug by centralizing the PERFORMANCE_LOG_PATTERN, extending regex coverage to support diverse log formats, and ensuring consistent metrics collection in container_runner, SLURM, and Kubernetes deployments. Leveraged Python and advanced regex engineering to consolidate code and automate comprehensive unit tests, resulting in a single source of truth for log parsing. This approach improved the reliability of performance analytics, reduced debugging time, and provided clearer capacity planning signals, reflecting a methodical focus on cross-platform consistency and maintainable code quality.