Worked on backend reliability and performance improvements across the apache/celeborn and apache/spark repositories, focusing on Java and Scala. Developed a configurable retry interval for commit operations in Celeborn, using a ScheduledExecutorService to handle transient network issues and reduce unnecessary retries. In Spark, optimized the GetJsonObjectEvaluator by reusing a class-level ByteArrayOutputStream, which decreased memory allocations and garbage collection overhead during repeated JSON evaluations. Addressed a bug in Celeborn by aligning shuffle fetch failure handling with Spark, preventing redundant stage reruns. All changes were validated with unit tests, emphasizing concurrent programming, performance optimization, and robust error handling across both projects.