A real world thing I did this week (which unfortunately is proprietary so I can't share), was a distributed configuration system for our cluster. It uses the RabbitMQ library[1][2] and Acid-state library[3] to have a per-node config-DB and uses RabbitMQ's guarantees for delivery of updates to the nodes. Messages are serialised using Aeson[4], the Haskell JSON library. I implemented a vector-clock versioning system on top of the Acid-state store to allow the clients to do conflict resolution. It really was a joy to write (around 350 lines of code). The C# client less so, 1000s of lines of boilerplate.
I think there's something really impressive about haskell; I tend to have significantly more trust in the code I write because mostly I know the edge cases don't exist. 9 times out of 10 if your code compiles it will work. Although be prepared for the compiler to hate you when you first start. The beatings are worth it though.
There's some basic code samples in the links:
[1] https://github.com/hreinhardt/amqp
[2] http://videlalvaro.github.io/2010/09/haskell-and-rabbitmq.ht...
Git-Annex https://git-annex.branchable.com/
All the tutorials and theory make sense, but making the jump to what I can do with other languages still seems a big one.