Your application is very interesting but it looks like you're trying to do too much at once. Especially with the AI part haha. I think a subset of it is achievable by one, focused person for a limited set of students and topics. As another said, the crap people deal with today is mandated by law and policy under a regime that won't buy into what you're proposing. Further, many places require by law that the kids are in what educational system considers a school rather than your app. So, you might want to prove the idea by hitting the market for workforce training and personal improvement then pivot into general education as both will use similar resources.
Another problem with your proposal, a huge one, is that there's nothing in it about countering spread of misinformation. Quick link to give an idea of how easy invalid information spreads and persists across social media:
http://www.iflscience.com/technology/facebook-echo-chambers-...
That or another Facebook study also pointed out that retractions (eg a Snopes article) only slowed misinformation in the 30-40% range IIRC. The bad info spreads more and lasts longer in your model. It has to be detected somehow with official counters on any topics where there's a consensus that they need to know. That means your model, like the others, needs human beings that act as sort of teachers and moderators. This is even true for a high-quality forum of smart people like Hacker News. Alternative is some AI that knows everything about how people speak, these topics, how to explain them... artificial teachers of nearly human intelligence that nobody has built yet. Since you lack that and need humans, your model can't scale for general population at the rate of technology scaling. Instead, it's something like N teachers per M groups where the N to M ratio is however much stuff the teachers can follow and moderate. You then modify this somehow with the fact that specific people or groups vetted by teachers as knowing the material can be rated among the N as their input is trustworthy. Might even compensate volunteers somehow a la StackOverflow or Quora if they contribute a certain amount.
Far as full experience regardless of bandwidth, do look up the concept of progressive enhancement as illustrated with this project:
https://gdstechnology.blog.gov.uk/2016/09/19/why-we-use-prog...
That thing even works on Lynx. Now, you want images, video, interactivity. The images can be compressed quite a bit while still being clear. Same with video. More compression for lower bandwidth links. Animated GIF's done well can illustrate in a few images what otherwise takes a video. Still popular on many sites. Use straight-up text where you can as my old text & RTF files were way smaller than Word or Web files. :) Support for distributed, file sharing within group if on Wifi. Alternatively, on SDcards where they could just buy one, have one person who affords Internet pull the data, and pass the shared card(s) along to group members.
I like how you mix it being a paid service with tax deductions and donations. I predict you'll probably put most or all of the services on the same infrastructure. What you can do is use the money that comes in for business or professional courses to cover creation of general ed material. The cash flow is split to cover each. The people making general ed material should have a track record of writing books or doing online courses that actually worked. They'll adapt it to your model. They'll have to change it for increased effectiveness as user feedback comes in like any other marketing. The cash flow also expands the distribution network so people with low income or poverty can access the material without overloading organization. A number of them might become contributors or paying users later.
So, those are some ideas I got as I read your application.