Worked on enhancing data integrity and metadata handling for PgVector upserts in the phidatahq/phidata repository. Focused on backend development using Python and SQL, addressing a critical bug to ensure upsert operations respect explicit Document.id values and reliably fall back to content_hash when necessary. Consolidated record construction into a dedicated helper function and improved the merging of filters into meta_data, while maintaining separate storage for filters. These changes increased the reliability and traceability of vector-based database operations, supporting more accurate downstream analytics and safer data merges. The work demonstrated attention to correctness and robust handling of vector database workflows.