Uses the Yesod web framework which uses a Haskell backend with its own 'Shakespearean Templates'. It's a bit outdated (stuck on Bootstrap 3) but it does do some clever stuff. The main advantage of Haskell + Yesod is that a lot of things become compile time errors instead of runtime errors. There are some disadvantages such as: limited documentation, quite painful to compile on a Raspberry Pi, and having to deal with monads to do stuff which would be simple in Python. But it's more fun than a React project.
Neural network predicting prices of used mountain bikes trained on 300,000 bikes scraped from the internet. Used inside an Android Kotlin app using Jetpack Compose. More information coming when I get round to writing it.