As a Google Professional Machine Learning Engineer at a startup utilizing Google Cloud, you are tasked with optimizing the data pipeline for multiple TensorFlow models. The data is in Parquet format, and the team requires a solution that efficiently manages this data both as input and output, without incurring additional infrastructure costs or complexity. The solution must also ensure seamless integration with TensorFlow to maintain development velocity. Considering these constraints, which of the following solutions should you implement? (Choose one correct option)