Worked on enhancing the ASR pipeline within the DS4SD/docling repository by developing a feature that allows the pipeline to accept both absolute file paths and BytesIO streams as input. This solution involved writing BytesIO contents to temporary files to ensure compatibility with the transcription model, while implementing robust error handling and automatic cleanup to prevent resource leaks. Leveraging Python for backend and file handling, the update improved the flexibility and reliability of content ingestion, enabling smoother integration with diverse data sources. The work focused on pipeline development and error management, supporting more versatile transcription workflows without introducing new bugs.