Worked on the NVIDIA/nv-ingest repository to deliver a comprehensive upgrade of the OCR model, replacing PaddleOCR with CustomOCR throughout the core ingestion components. This involved refactoring extraction logic, updating model interfaces, and revising configuration schemas to support the new backend. The transition required renaming functions, adjusting method calls, and expanding unit tests to validate the stability and performance of CustomOCR. By reducing dependency on PaddleOCR and enhancing maintainability, the work established a foundation for future OCR improvements. The project was implemented using Python and leveraged skills in API development, image processing, and unit testing to improve model flexibility and performance.