I don't get how anything designed in the past couple years for something like this could be done in such a way as to crash and burn like this...

I mean, a couple geo-redundant C* clusters, and a load-balanced API server... Hell, design around (AWS|Azure|Google) cloud's big-data offerings, provided they meet legal/security concerns, and this would have been highly unlikely, or at least had appropriate help from the cloud provider for what that structure would have brought.

I don't know the details, but a multi-million dollar project for something like this should have been able to handle a LOT of load, if properly engineered like a project such as this should have been.