Your data science team is tasked with conducting rapid experiments involving various features, model architectures, and hyperparameters. They need an efficient way to track the accuracy metrics of these experiments and access the metrics programmatically over time. What approach should they take to achieve this while minimizing manual effort?