Worked on the volcengine/verl repository to address a critical bug in the DataProto module, focusing on improving the reliability of data indexing operations. Tackled a type-related error in the select_idxs method that previously caused incorrect batch size handling when using boolean and integer indexing. The solution involved refining Python typing and implementing comprehensive unit tests to validate select_idxs across a range of input types, ensuring robust regression protection. Leveraged Python programming and data manipulation skills, along with git-based traceability, to deliver a targeted fix that reduced the risk of downstream runtime errors and enhanced the maintainability of data workflows.