Worked on stabilizing the DOCX ingestion and embedding pipeline for the infiniflow/ragflow repository, focusing on backend reliability and data integrity. Addressed a recurring issue where whitespace-only content from DOCX files caused errors in the Zhipu embedding API by implementing a sanitization guard that replaces such content with 'None' before embedding. Used Python for backend development and data parsing, validating the solution against large, complex DOCX files to ensure robustness. Collaborated closely with another developer to co-author the fix and performed comprehensive end-to-end testing, resulting in reduced data processing failures and improved reliability of the embedding workflow through effective API integration.