Andreas Wieweg: I’d probably want to add a little bit of the Shell use case.
Sramana Mitra: I would like to understand how it’s built together. What are the data sources? How does this get implemented from a technical point of view?
Andreas Wieweg: In terms of data sources, in my experience, they vary a lot in different enterprises. One might think that most enterprises have very modern, well organized, structured data sources. Actually, some do and some don’t. In this case, data comes from various data sources. They are not always accessible in APIs. If they’re accessible in APIs, we love that and we just communicate with APIs. In other cases, data sources might live in a database.
We use various technologies to input those sources. In this case, there are different data sources. We use data sources both to build the various entities and to interrogate them. Basically, we have the ability to interrogate from an API level to a loose bunch of data.
Sramana Mitra: What is the AI algorithm? Can you talk about whatever is convenient to explain how the algorithm works?
Andreas Wieweg: First I just want to get back a bit to the point that we have a full coverage platform. There are a lot of different algorithms in our platform. This is a platform that keeps the user covered from the inception of the application. We have components on the platform, which allow you to analyze the data, make learnings about that data, and find what is in the data. It then goes to our build environment to develop the application. You then launch it in our engine.
Sramana Mitra: Let me ask you the question differently. I understand that you have speech processing algorithms. First and foremost, you have speech recognition and then it understands what’s the question. Then the various data sources have been brought together to train your algorithm to be able to answer questions with vocabulary that is specific to the domain of the enterprise that you are working with.
Are you using some sort of an expert system type of technology? What is the general direction of the interpretation and learning that is happening?
Andreas Wieweg: We do not have speech component. That is not in our platform. We are agnostic. We use various speech-to-text providers suitable for an application. Our platform handles the text which comes in.
Andy Peart: You can think of this as three components. If you speak to Indigo, a mobile-based personal assistant, you need to pick up the analog sound file and do something with it. It comes in to Teneo, which is the brain of the device. It works out what to do. First, it will analyze the import. It will reason as to what to do with it. It will either ask back a question or look up information from an enterprise system. If you ask about the weather in San Francisco, it will go to a web service to give you an answer.
Sramana Mitra: I got that. I’m trying to understand the technology of how this is implemented. Is this some sort of a rule-based engine that has all sorts of different questions and possible answers? What is the heart of the algorithm?
Andreas Wieweg: At the heart, you have what we call a collection of dialog flows where you have designed the dialog behaviour. Those dialog flows are executed by our engine. It’s not an expert system but it executes and runs these dialog flows.
Sramana Mitra: It operates as a rule-based system essentially. You have all sorts of dialog flows and it operates as a rule-based system based on what it’s interpreting.
Andreas Wieweg: Yes and no. That is your dialog behaviour. Then you have the understanding or mapping of the natural language. That might not necessarily be rule-based but you may have machine learning algorithms. It’s a mixed bag of technologies.
Sramana Mitra: You’re taking the text questions in natural language, and then you’re trying to interpret those questions and map it to some dialog flow.
Andreas Wieweg: We match that input to the context of the dialog flow.
Sramana Mitra: I got it. My next question is what is the adoption level of the kinds of technologies in various large enterprises. Is this happening at a fast pace at this point?
Andy Peart: I think there’s no doubt that it’s still at an early stage overall from a market point of view. It’s not yet at the stage where people are just, “I need a whole financial system. Which one shall I go for?” The whole marketplace that we are operating in is still in the fairly disruptive stage.