What led me to move into AI consulting

Despite having a broad understanding of various domains, solutions, and technologies, I felt that I lacked an insight of the systems I engaged in developing. That’s why I began my serious study of adtech both as a business and as a set of technologies used to automate and optimize business processes and improve competitiveness. This was when I first encountered the concept of big data and content personalization.?

Blockchain was the next area that interested me at the time, just when everyone was starting to talk about it. I studied the technology and its potential applications on a basic level to gain a general understanding of its potential. It didn't seem ready for real business application at that time to me, and my interest waned.

Afterward, cycling through areas of innovation, I studied the basis of both data science and machine learning and their applications. The results of my studies surprised me, as it became clear that AI held great promise. I went deeper into my studies, completed numerous courses, participated in thematic communities, and attended online training sessions. At that point, NIX was gaining interest in AI on the corporate level, so it was essential to undertake more studies and arrange the system of work in this new direction. During that period, I became involved as an AI researcher and consultant.?

My first project in my capacity as an AI consultant was a customer support system to automatically assist airline passengers who had problems with their flights — such as the flight being canceled or rescheduled, or the passenger being late or missing the flight. Following transportation failures, different kinds of compensation are due, and it is quite challenging for transport operators to manage these requests on a mass scale. As a result, we came up with a chatbot that would automatically answer common passenger questions and direct them to additional information resources.?

The initial level was a semi-automated system that required operator confirmation of replies and was only available in English. We continued to work on optimizing the system after we met our planned outputs, as well as porting it to other languages, which was a new challenge. We realized that the practical implementation of well-known methods always deviates significantly from theory and that each instance has its own set of unknowns and many exceptions.

While writing this post, I was reminded of other interesting cases that I’ve decided to cover in my next post.