During their work on the facebookincubator/velox repository, Dnb developed a memory-efficient approach for constructing CudfVector objects from cudf::packed_table, focusing on scalable data processing. By leveraging C++ and GPU programming, Dnb avoided unnecessary data copies and enabled table view-based exposure, which reduced memory usage and improved throughput. The technical solution involved refactoring operators to operate directly on table views rather than materializing full tables, resulting in performance gains for data processing workflows. This work demonstrated a deep understanding of data processing and memory optimization, delivering a well-integrated feature that addressed both efficiency and scalability within the Velox project.