Initially, he developed a system for drone camera stabilization, then participated in designing the “air” display, and later on, founded a company called EORA offering solutions based on artificial intelligence (AI, Data Science, Data Mining). His name is Roman Doronin.
Roman became a PhD in Economics of Innovation at M.V. Lomonosov Moscow State University and was a four-time winner of the award “Our Moscow Region”. Recently, his EORAcompany has won the contest organized by the leading payment service provider, QIWI, and is currently implementing a project for it.
We talked to Roman Doronin and discovered how he started his career in the AI sector, about the competition in this niche, and what challenges could be faced while working with high profile customers.
My activity was not quite related to artificial intelligence at the beginning. First off, I worked for Sviaz-Bank as an information protection specialist. At that time, I studied at Moscow State University of Geodesy and Cartography at the Faculty of Applied Cosmonautics and worked alongside.
“We became one of the first Skolkovo residents”
My first project started in 2009. We designed a camera stabilization system for unmanned aerial vehicles, preventing images from shaking during flights. Afterwards, I co-founded one of the best known startups by that time – DISPLAIR designing an “air” display.
We became one of the first Skolkovo residents, attracted huge investments, and were a well-known company within four years. We were introduced to Dmitry Medvedev and other high-level public officials. Besides, we became residents of the IT Park inTatarstan. Thereafter, I focused on EORA. However, the company did not initially design artificial intelligence solutions.
We produced a service, which was the so-called tow truck Uber. had good expertise in combinatorial optimization, i.e. in building the best possible routes, and thought that there were enough technologies to establish successful businesses. We renamed the company as EORA.HELP. It still exists.
“We started with such technologies as natural language processing”
The company sells packages of roadside assistance services. A package is bought as an insurance: it includestowing, car assistance, legal advice, and fuel supply. All of this is valid over the year.
Later, we decided to create EORA.HELP’s assistant. We took interest in this area and explored chatbots. This resulted in the establishment of a large company – EORA: solutions based on artificial intelligence. It was in 2015 – 2016; quite a long period, but such a thing could not happen in a single day. We started with such technologies as natural language processing (NLP). A part of the whole dialogue system is based on this technology.
“The team is united by hackathons”
The company entered this sector and took part in programming competitions and hackathons. Currently, our team is a frequent winner of international and Russian programming contests. Gradually, we were joined by specialists in other AI areas, for example, computer vision and Data Science.
Now, our group consists of two companies. One of them, EORA – a dialogue system, is responsible for communication automation. The second one, EORA Data Lab, is engaged in Data Science and computer vision.
“There is no general recipe of success”
Our advantage is that each company focuses on a certain activity: one is in charge of language and text, and another on Data Science and computer vision.
The speed is also our advantage. We rapidly implement projects because of having a lot of best practices. Moreover, we are profitable in terms of cost, as we are based in Innopolis and can afford lower prices than average prices on the market. We always learn something new by participating in almost all the hackathons. We solve new tasks week by week.
To find specialists, one should be in the mix. Therefore, we should be involved in hackathons and specialized conferences, or attend the Yandex Data Analysis School. Nevertheless, there is no general recipe of success. You should just act a lot, and it does not necessarily mean that you will find strong specialists, because major companies are ready to offer them favorable conditions.
“The bigger company, the more pay-off difficulties”
The hardest part about working with major customers is the coordination of contracts, which can last up to six months. And the long process of post-payment: this period is 60 days in many companies. Thus, challenges likely belong to businesses rather than to task solutions.
Everyone wants to cooperate with large corporations, but be set to the fact that the contract will be very long. You will be driven into a corner in terms of deadline and money, while conditions will bechanged on the fly. You should especially take into account a long payment period: small companies frequently face cash deficiencies, because the order is fulfilled but the customer has not paid for two months.
“I do not see strong competition, the market is almost empty”
The bigger company, the more pay-off difficulties. One should always have a fiscal buffer in such a case. Nowadays, many experts say that one should withdraw net money from the company and, for instance, buy cars, but I consider this the height of stupidity. Do not withdraw anything: the main thing is to have a buffer for three-four months of autonomous operation. So, you should focus on accumulating this buffer. Extra sums can be invested in the company and new employees, as well as in sales.
I do not see competition for us. Besides, there is poor competition in this sector, because similar solutions are quite rare. Firstly, finding a competent data scientist and computer vision specialist is a pretty difficult task. Secondly, customers often do not know what they want, and you should develop skills and case studies in this area. Thirdly, it is just the beginning of the big age similar to the emergence of websites and the Internet. Today, few people know how to work with this efficiently.
“We give machine learning lectures every week”
Our company has a social project called AI Community: every week, we give lectures on machine learning, computer vision, and Data Science, as well as reveal hackathon case studies. Lectures are delivered by all my partners and EORA co-founders. We have been conducting these activities at Innopolis for more than a year.
“We want to get data under control”
One of the most interesting projects we are implementing is data analysis for QIWI. QIWI’s business is based on processing activity, i.e. transfer of payments. Huge financial sums estimated at billions are sent through the system. Sometimes a small part of these transactions fails to pass.
Our project is aimed at getting data under control, examining customer behavior and its structure, finding the reason for some transactions failing to pass, and eliminating these causes. Roughly speaking, if 50 million payments out of 1 billion fail to pass, fixing a certain technical error, we will help QIWI to save a huge amount of money, allowing to work seamlessly.
Now, we have passed the halfway point of the project. Its release and financial results will be available at the end of December. At the conference, we will present the first figures as well as describe what we have found, how we have worked, cleaned data, and provided the speed.
Roman Doronin from EORA will make a collaborative presentation with QIWI’s analyst Dmitry Zamaleev at AI Conference. The event dedicated to the development of artificial intelligence technologies will take place in Moscow on November 22.