Worked on stabilizing and enhancing data workflows in the meta-llama/llama-cookbook and huggingface/transformers repositories, focusing on reliability and maintainability. Addressed notebook environment issues by refining dependency management and updating data preprocessing paths, ensuring Jupyter notebooks ran smoothly with current data inputs. In the transformers repository, improved model robustness by correcting output structures for video classification and adding explicit error handling for chat templates, accompanied by expanded unit tests. Leveraged Python, Jupyter Notebook, and deep learning frameworks to deliver targeted fixes and updates, reducing runtime errors and supporting more reliable onboarding and downstream product integration across both projects.