curious how this scales with larger datasets. anyone tried it in production?