Worked on the Tencent/WeKnora repository to enhance the document parser, focusing on preserving the original MinerU Markdown structure and improving the handling of relative image references during content conversion. The approach involved targeted normalization of Markdown and HTML, extending the ImageResolver to treat relative HTML img tags consistently with Markdown images, and deduplicating redundant saves to optimize storage. Developed and integrated focused tests to ensure regression protection and maintain compatibility with existing features. Utilized Go for backend development, along with HTML and Markdown parsing, emphasizing storage pipeline normalization and test-driven development to deliver higher fidelity rendering and improved maintainability.