So just read and judge every piece of text entered into the whole platform, 24/7, 365.2425 days a year, both in and out of context.

> So just read and judge every piece of text entered into the whole platform

No. Not at all.

We're software engineers, right? Make the machines do the heavy lifting.

Create a simple keyword based alerting mechanism for known inflammatory language in the languages that you support. There are resources to help you make this easier in multiple languages, i.e. https://hatespeechdata.com/

Add a moderation system for your users to flag abusive content.

Until you grow to a huge size, or unless your platform is courting violence promoting content, there should be a relatively small amount of content you need to manually review and remove.

twitter did this, IIRC they needed to turn it off as it was flagging almost all republicans

I seriously doubt that Twitter has no system that internally flags potentially abusive or violent tweets using machine automated classifiers. This stuff is table stakes nowadays for any major service that accepts user generated data.