4.2 gigabytes???? That's insane, especially considering how poorly optimized the models are (with better algorithms and ways to understand and categorize the data, this could be easily reduced 10x or even 100x).