Vowpal-Wabbit powers our online-learning prediction engine, that suggests our users replies to send out, and labels to add to their data, based on previous behavior by that user.
Used as central Message Broker; off-loading tasks to be executed asynchronous, used as communication tool between different microservices, used as tool to handle peaks in incoming data, etc.
All social data we track is stored in ElasticSearch to make it easily searchable and for advanced statistics. Our several ElasticSearch clusters hold several billion social messages.
Used as a caching layer (when we need more functionality than simple key-value storage); keeping lists of online users, used for our smart-assigning feature, keeping track of sliding-window rate limiting information.