Worked on the anyscale/templates repository to improve the reliability of Hugging Face dataset loading within Ray Data workflows. Addressed a recurring issue with the cnn_dailymail dataset by implementing a workaround that replaced the unstable from_huggingface call with ray.data.read_parquet, leveraging HfFileSystem for seamless integration. This solution enhanced the stability of the data ingestion pipeline, reducing failures and simplifying debugging for users working with large-scale text datasets. The work was carried out using Python and Jupyter Notebook, drawing on expertise in data engineering, Hugging Face Datasets, and Ray Data to deliver a more robust and maintainable loading process.