Worked on the infiniflow/ragflow repository to enhance the stability of the OpenAI chat completions input pipeline by addressing message normalization challenges. Focused on processing arrays and mixed content types, the solution involved concatenating only text segments before further handling, which reduced server errors and improved compatibility with OpenAI-compatible clients. The work included fixing the API endpoint to prevent TypeError and HTTP 500 responses when non-text content was present, aligning normalization with downstream dialog service requirements. Utilized Python for API development and unit testing, laying the foundation for broader content-type support and more robust error messaging in future updates.