As I can’t log in, I can’t tell if Elixir is needed on the backend. If the backend is trivial enough it can easily be done with firebase (+cloud-functions) or sth similar. Elixir shines brightly on the backend when it gets a little more involved.
What don’t you like about flutter-web? Isn’t that just a “normal” flutter app compiled for web?
For a MVP/POC I can’t imagine anything more productive than LV, just throw it away if you think it works (or keep the backend maybe).
The way the compile their code to JS leaves bugs that are really hard to hack around. For example when typing, depending on the component the text in the input will move up and down. Took me a good 3 hours for a work around. This is due to carat rendering and so you need to shrink the carat. That’s just one off the top of my head.
FE/mobile dev is hard. You say the primary pain is the split but for a MVP I would think the Elixir side would be pretty frictionless. Maybe my perspective is skewed because I am primarily BE but I think the main problem with most indie app launches is the design/UX, because it takes a lot more resources to get that right than BE development does. Looking pretty, smooth transitions, etc are all critical.
From the POV of an informed user, rather than a dev, if I am testing out an app and it’s ugly/clunky, I am inclined to throw it out right away, even if potentially offers a lot of value in features. Conversely, if I am trying out an app that looks and feels great, but 3/4 of features are “pending”, I’ll hang around as long as it is adding some value. If I try to do something and it just errors out, but I get a slick notification and no FE crash, I sleep. Some action actually works but the page is ugly and I can barely tell it did work? Real ****. No thanks. And this is 10x as true for mobile apps as web apps.
So I guess my advice is, whatever helps you get the best looking UX as part of the MVP, do that. Skeleton prototypes really only work for internal purposes (even then you get complaints from colleagues 
)
Any web app in any programming language can be a a PWA, but PWAs cannot be uploaded to Android or Apple stores, because they are just web apps. A web app just needs a manifest.json file to instruct mobile devices how to add it to the home screen to then open them like if they were a “native” mobile app. So, this is all you need to have a basic web app working on a mobile device as a PWA app. Off-course more can be done to leverage the PWA concepts in a web app.
Supabase is built with Elixir but they have different client libraries with for example Typescript. So the backend Supabase is running is in elixir (meaning you get many of the good stuff from the beam) but you interact with it using client libraries like typescript. So the backend is elixir but yea you don’t really see any of the elixir.
I see a “realtime” repository using elixir, but I do not see a core repository with Elixir ( Supabase · GitHub ) so I’m not really convinced.
They use a whole slew of technology building on top of postgres, but afaik realtime is a core part of their offering. But yes it’s not a big elixir monolith + postgres.
This has a bit more on what they use elixir for: Thinking Elixir Podcast 73: Elixir at Supabase with Paul Copplestone
I have not seen this mentioned above:
Use react native with some way to share code between JS runtime and backend.
Because personally I think JS and TS are solid as shredded salad I did some research on typed languages that compile to JS but provide strong guarantees of error free code.
I looked at Haskell + Purescript, Ocaml + Reason/Rescript but in the end I decided to give Gleam a go. It is less obscure than Haskell and in better community shape then Reason/Rescript. Aaand I could share gleam code with an elixir backend.
I only started, and now have a react native app with main engine written in gleam and compiled to JS. The views are written in JSX and have minimum logic (I try to put it in custom hooks)
I hope to post a longer write up on how it’s going, so far I have hit two issues: