Worked on backend stability and metadata reliability for the ragflow repository, focusing on the OpenSearch integration. Addressed two critical bugs by implementing the missing get_scores path for KNN second-pass scoring, which prevented AttributeError crashes and ensured stable retrieval operations. Improved document metadata handling by normalizing dynamic mapping and aligning search parameter names, eliminating mapper_parsing exceptions and TypeErrors. Harmonized the API surface between OpenSearch and Elasticsearch backends by standardizing parameter naming conventions, which preserved compatibility and reduced integration friction. All changes were isolated to OpenSearch, maintaining backward compatibility and establishing a clear test path. Utilized Python, OpenSearch, and Elasticsearch.