How to communicate with people is not something easy and I realized it even more when I had, with my team, to teach a robot to talk with each other!

At BNP Paribas Fortis – Home for Innovation – we had the choice to build a language thanks to big vendors tool or to build it with an open-source technologies.

The choice we’ve made was based on building word vectors from scratch in order to be in a perfect match with the bank reality and financial terminology. In addition, by choosing an open-source bot, we had the opportunity to build our own chatbot in many languages without having to wait for a latest release from top chatbot vendors, mainly for Dutch in our case. What helped us during this step, for choosing the right solution, was Cream Consulting’s approach, which allowed me as a Creamer to let agility works inside the Dev Team.

Our objective was to build a smart conversational textual chatbot to answer basic repetitive questions. This repetition is like teaching a child to talk. The more you repeat the same words, the more your bot is learning to use the appropriate words in a given context. Different contexts, different words and your bot should be able to converse with an individual of flesh and blood, without this one could be able to determine if, somewhere on the other end of the terminal, there is a human or a robot. And trust me, as a human, I was sometimes bluffed by the ability of a robot and the tenacity of our work to obtain this result.

The challenge was to give the opportunity to the bot to converse with multiple individual to allow him to gain in subtlety. The most important is the conversations database, the greatest is the chance for the bot to find an appropriate statement. And then, the real turning point was to develop an evolutive chatbot, identified as « learning AI », with a new definition of intelligence: go from a reactive vision logic (already given information) to a predictive emotional vision (research of appropriate response), closer to a real human sensitivity.

Some methods help often to tackle a challenge, and that’s where my experience in Scrum was so beneficial; helping to organise, prioritize, test developments and progress at each iteration.

We tried to give our bot different personality traits based on identified Personae in marketing and commercial contexts, which is more than a challenge in terms of sensitivity.

Welcome to a world where the user discusses with all previous user ghosts, brought together in a single entity.