In the second, I'm using a bit of Datomic mainly to save UI-related and small bits of user state. Mainly I use Cassandra, Redis, Kafka, and Elasticsearch. I should add in this project, I'm also using untangled.
I communicate a lot over websockets as well, so between that and Cassandra I have to do a lot of mapping and transformations on my own, however once you get it setup the rest is mostly magic. Most of my reads and writes over the socket are happening with consumers and producers to Kafka so mapping om-next data here is just standard Clojure, nothing fancy.
When I do query Cassandra directly, the queries in particular by nature need to be somewhat restrictive which it turns out play nicely with om-next since mostly for reads I just need to worry about the "select" columns. Redis I just use for LRU caching and to buffer some video-like blocks of data before the user tries to access them. Elasticsearch is just used for searches and some type of filtering, but gets populated through Cassandra data anyway so again no problem with om-next.
I'd say the most annoying thing here was as I hinted, getting it all setup. Things are clearly built for Datomic, but there are integration points all over the place and just using standard functions in Clojure to manipulate vectors and maps is easy. If you're not experienced with datalog or datomic pull syntax, things might be a bit harder at the start for some people.
Regarding API calls, I should add since I am using untangled, I use the default websocket lib which is just sente. I have rigged things up with aleph and yada, but also played around with standard compojure api and other offerings. Long-term I'd like to stick with yada, but I've encountered some documented bugs that are annoying so I've put my focus elsewhere in the short-term when needed.