Kun De Wan focused on backend stability and observability improvements in Apache Spark and Gluten over a two-month period. He addressed critical bugs in the Spark repository by defensively filtering null MapStatus entries during shuffle data cleanup, reducing the risk of NullPointerExceptions and improving job reliability for workloads using the ExternalShuffleService. In addition, he enhanced query execution stability by updating HiveGenericUDFEvaluator to skip FoldableUnevaluable expressions, preventing runtime exceptions. His work in the Gluten repository improved shuffle time metrics accuracy for Celeborn by refining metric selection logic. He applied expertise in Scala, SQL, and performance tuning to deliver robust, maintainable solutions.