Worked on the DS4SD/docling repository to enhance the reliability of the ASR processing pipeline by addressing FFmpeg dependency issues. Developed robust error handling in Python to detect when FFmpeg is unavailable, ensuring that users receive clear, actionable error messages rather than encountering silent failures. This approach allowed the system to halt affected audio processing gracefully, reducing troubleshooting time and improving maintainability. Additionally, optimized unit testing by moving common imports to the top level and implementing lazy loading for resource-intensive model imports, which improved test feedback speed and reduced startup costs. Focus areas included audio processing, error handling, and test optimization.