Retrospective of the year

This year we had the stage of onshoring our artificial intelligence technologies for business needs. Initially, we created a separate "simplified" version of the artificial intelligence core, which can be used as an assistant for customers and business employees.

Additionally, we had to solve three difficult tasks. The first is to enable customer employees to train AI themselves and monitor its work. The second is to organize the output of information to various communication channels. The communication channels differ greatly in the possibility of using interface elements in the communication process. The third is to create a single assistant working with both clients and employees.

The output of information should be limited to the roles of the interlocutor. So, to the same question, the assistant's answer may vary depending on the role of the interlocutor. These tasks were successfully solved. A software package was built around the assistant, which we called Steos. We implemented it to the first customers.

When we were ready to move on, a new problem arose – we need to develop two AI cores: separately for business and separately for scientific developments in the field of AGI. During the year, interesting solutions have accumulated in each of the cores that can be applied in both directions. It was decided to combine two AI cores into one. And since the core is supposed to be used in the mode of simultaneous work with a large number of interlocutors, we decided to rewrite it in the Go language. We believe that this will make the most efficient use of the resources of the devices on which the AI will be launched.

Additionally, we have revised and worked out the knowledge models and made minor changes in their structure. We have developed tools that allow even non-specialists to work with knowledge. This will speed up the addition of new knowledge by several times.

The new unified version of the kernel is ready and used in commercial applications. We managed to achieve a serious increase in productivity. So, the time for analyzing the question and synthesizing the answer decreased by 30 times.

Memory usage for the basic core structures has decreased by 5 times. AI can now be used as a microservice, being built into various software complexes.

Several installations have been carried out to customers and now we are working on feedback to improve the products.

Also, this year we have made a lot of progress in speech technologies. The speech synthesis we have created is already one of the best in the world. This work will be described in another article.