What a bummer to have to maintain a voice assistant on top of all the ways you are currently talking to your customers!
Recently, we came across a project where the customer had a very large and well-defined website and the prospect of maintaining a voice assistant as a parallel content stream seemed daunting. It would have been great to use a few APIs to pull everything together in a dynamic way, but this customer, like most, didn’t have the relevant APIs.
We came up with an interesting solution – scrape the website content in a way that we could extract question and answer pairs – whether they were from an FAQ list or from a list of titles and paragraphs. Apply some supervised machine learning to this and build for the various voice platforms the customer wanted to serve.
The main benefit is that the customer only then had to maintain one content source, their website, and when the content was formatted properly, could be used to create a voice assistant – automagically.
Here is the logical diagram: