Worked on the envoyproxy/envoy repository to optimize the Prometheus stats endpoint, focusing on backend performance and system design using C++. The project targeted reducing scrape latency and memory usage for large metric groups, enabling more efficient observability at scale. The approach involved replacing an rb-tree based grouping structure with a flat hash map, implementing in-place sanitization for metric names and tags, and modifying snapshot data directly to minimize memory allocations. These changes resulted in a 30-40% reduction in scrape latency and improved memory efficiency, reflecting a deep understanding of performance optimization and low-level backend engineering in C++.