Worked on the pinterest/ray repository to enhance data ingestion workflows by implementing selective fragment reads for Lance datasets. Leveraging Python and data engineering expertise, introduced scanner_options to ray.data.read_lance, allowing users to specify which fragments to read and thereby reducing unnecessary IO and memory consumption. Developed comprehensive tests to validate fragment selection and ensure regression safety, while also addressing a known issue with fragment handling to stabilize the feature. This work improved the reliability and efficiency of distributed data loading for Lance-backed datasets, supporting more scalable analytics pipelines and contributing to cost-effective resource utilization in large-scale data engineering environments.