Developed memory-efficient CudfVector construction and view-based optimizations for the facebookincubator/velox repository, focusing on scalable data processing with reduced memory usage. Leveraged C++ and GPU programming to introduce support for constructing CudfVector directly from cudf::packed_table, eliminating unnecessary data copies by exposing table views instead of materializing full tables. Refactored operators to operate on these views, which improved throughput and lowered the memory footprint during data processing tasks. The work centered on enhancing performance for large-scale data workflows, demonstrating depth in C++ development and data processing, and was delivered as a merged feature without any reported bug fixes during the period.