The voice recognition is good, but the interface is completely independent of Alexa. The interface is spawned from a group of people brainstorming all the diff

The voice recognition is good, but the interface is completely independent of Alexa. The interface is spawned from a group of people brainstorming all the different ways someone can ask to turn on the lights, and then figuring out the ways to deal with all the different inputs. I'm not sure how common the knowledge is around making an Alexa skill, but I made one as a side project so I might have some insight. You define a bunch of variables, and describe what form they are likely to come in (number, data, US city, list of strings). It's not strictly bound to any enumeration you use. Then you list out a bunch of phrases that people are expected to say to activate a function. You can pepper the variables into these phrases. Then you link these phrases to functions, which receive the variables as parameters. What you do from there is up to you. I would say it's a clever way of gradually reducing the complexity, but it does result in putting the "conversation" complexity on the developer. There is no "state" or compound queries unless the developer thought of the conversation going that way.