Presentation: Serverless and Chatbots: A Match Made in the Cloud

Gillian: I work for a large enterprise organization, Liberty Mutual. About a year and a half ago, I started working on an employee digital assistant (a chatbot for internal productivity). That was the first time I'd worked in the cloud. Suddenly with this project, I was working with cloud, serverless, and artificial intelligence.

While I was working on these things, I started to feel like there was a much, much bigger picture at play here. We are a huge company, and we've got all these different initiatives going on. They have so many commonalities. Not only that, but because it’s a new interface, it's just different than anything we've ever built, and we have all these new things to think about (and learn). It’s not just in IT but outside of it as well.

So, at the moment, I've stepped out of the project (just briefly) to put together an organizational strategy for how we can equip the company to use these technologies and be really effective.

Gillian: I'll be speaking about chatbots generally, and how we built our employee digital assistant. I will be focused on building on AWS. That's not an endorsement of AWS. That is just what we built our platform on. A lot of the learnings however are applicable regardless of what cloud or chatbot technology you use.

As far as technology, our app is surfaced through a toolbar on our intranet as the primary way of interacting with it, and on our mobile apps, it has a custom front end. The backend is designed to be able to surface up through anywhere. My talk won't be focused on the front-end integration, it will be more focused on the backend.

Gillian: We're using AWS Lex, which is their service for conversation bots. We're using a lot of the services from AWS as well. That's the power of serverless. There are no servers but you just put together a whole pile of services to put your building blocks together. We are, however, using our own CI/CD tools outside of AWS and build everything through code not the console.

Gillian: I will show a very high-level diagram of our architecture, where we say things, at what point we say things, and how we do our conversational analytics (things like that). It won't be huge, but I will show the various bits of the architecture, and what happens in those different bits. I’ll also talk about our build/deploy pipeline and how we test.

Gillian: I'm talking to an experienced developer, but probably someone who is just getting started with chatbots (potentially hasn't even done anything with chatbots yet). If you're getting started (or you're wanting to scale up something you've got) and you're trying to work out what you need to think about if you're going to go to scale, this talk is for you.

I will be assuming that you know about software development. so I’ll be jumping right into the pieces involved in building a chatbot, and why we used serverless.

Gillian: Because I work in cognitive technologies and think a lot about humanizing technology (what we should be doing), I get a lot of questions around what human capabilities I believe we can or can't automate now or in the future. But honestly, that's completely the wrong question. The question people should be asking is what human capability shouldn't we automate? Regardless of whether we can or can't, which one shouldn't we?

That's a question that keeps me up at night. That's the one I think about a lot. Honestly, I think we'll be able to create a facsimile of almost anything, but there are things that even if we can,  we shouldn't.