Interview with Nikita Smooth: how artificial intelligence works, which you can trust

On April 14, 2025, Openai introduced a new generation of artificial intelligence models GPT-4.1. These models are especially good with programming tasks, large volumes of text and accurate instructions. Already in the early days, they aroused great interest among the developers – from -with higher quality answers and reducing the cost of use.

Against the background of such technological breakthroughs, it is especially important to hear the opinion of those who work with AI, not at the level of theory, but introduces it into real digital products. Nikita Smooth is one of these specialists. He created a solution using AI for banks and trading floors, and now works with Primer AI, where he was engaged in intellectual search engines. In 2023, he received the prestigious Braintech awards for his contribution to the development of applied artificial intelligence.

We talked with Nikita about how he came to the AI ​​sphere, on which projects worked, and, in his opinion, the technology awaits in the coming years.

From applied science – to scaling systems

– Nikito, your story in AI has begun quite impossible. Tell us where it all started.

It all started in 2014, when I worked at the research Institute of Hydroeuvity. There I created automated systems for measuring acoustic fields using Python, C and Matlab. This was my first experience in the junction of engineering, data and programming. But my admiration for the technology and the possibilities of their use in the real world led me to machine learning and artificial intelligence. Since 2017, I started working on projects related to automation and creating effective software solutions using AI to optimize business processes.

One of these key stages was my work in Lalafo – a market that worked simultaneously in several markets. There I headed the project to create ml-paying for image processing. We used streaming technologies – Kafka, Elasticsearch – and machine learning to analyze photos and accelerate content moderation. This made it possible to significantly improve work with great data and increase the performance of the system. I am especially proud of this project: the developed automation system has received international recognition and received the Brantech award for contribution to the development of AI. In the future, I continued to develop solutions for other companies such as Revolut and Primer AI, where I already worked with more complex systems and real data.

Breakthrough: A project that changed the rules

– You mentioned the Brantech awards, which is awarded for significant technological solutions in the field of AI. Tell us more about the project that brought you this recognition.

-I received a reward for developing a system of automation of intellectual content processes for international trading platforms Lalafo. Usually, the user had to manually upload photos of the product, fill out the description, choose a category, indicate the characteristics – color, size, status. It took a lot of time and created a barrier for advertising.

I have developed a system that makes it automatically. The user uploads photos of the product and model:

• recognizes the object (for example, t -frish);
• defines the category of goods;
• The main visual characteristics: color, size, condition;
• estimates the approximate price;
• And as a result, he forms a finished announcement – from the text to the recommended category.

This development made it possible to greatly simplify the process for users, accelerate content loading and improve the search algorithms on the platform.

– Your development has become an example of how artificial intelligence can be practical – to optimize processes and make technologies more accessible to users. Why do you think this project was so highly valued by the jury?

– I believe that the practical significance and maturity of the solution have become key factors. The system that I developed was not just a prototype, but also a full -fledged product integrated into a real highly executed ecosystem that really worked: reducing the load on moderators, accelerating publications, assistance and searching.

And most importantly, my project clearly demonstrated how AI can be useful without unnecessary “magic”, now for millions of users. I think that this real benefit for people and business has become a decisive factor in obtaining a high rating.

What was as follows: fintech and real tasks

– After the successful implementation of the project in the markets, as far as I know, you switched to new major tasks. What problems did you expect from you further?

-The next step was the work in the FinTech-I sector joined the Revolut team, where the task was even more difficult: real-time work, high requirements for reliability and data processing speed.

– Revolut is one of the largest European fintech. What does the work with AI in such systems are especially difficult?

Revolut is a bank with millions of users, and the system should be equally reliable. Unlike experimental environments, there is no error in FinTech. Where the system works with real data, the cost of delay or inaccuracies is real money, user experience, trust. In this company, I managed a project that processed the events of more than 100 sources and processed about 200,000 events per minute. This is not only banking transactions, but also any updates of user actions, system systems, interface interaction. The architecture should be stable, fast, scale and safe. This experience was very important – he taught me how to work with real time, with high trust systems, where there was no place for failure. All this was the basis for working with language models, already in Primer AI.

How to build a si who can be trusted

– Today you are working in Primer AI – one of the leading companies in the field of intellectual search solutions. What tasks do you encounter now and how do you solve the problem of trust in AI?

One of the most important tasks that I managed to solve is to reduce the errors of AI. The system sometimes “invented” the answers – answered with confidence, even if I did not know the exact answer. We have developed a special check, which we managed to reduce such errors by 500 times – from 5% to 0.01%. This significantly increased the reliability of the system.

Another direction that I work on is a search engine. Imagine: earlier she responded to requests in 10 seconds, and now less than one. This is especially important when the user is looking for something complex or urgent. In addition, we are looking smarter and personalized. The system begins to understand what is interesting for you – for example, given what topics you are looking for earlier, what you are doing, in what context you are asking questions. This allows you to give an accurate, adapted person, and not just a list of documents.

And most importantly: finding is not enough. It is necessary to turn “unprocessed” data – tables, reports, fragments of texts – into finished, clean and useful information. To make a person not just a set of facts, but also a clear, structured answer to his question. Currently, I just make the construction of the following systems: from the request – to the point, from data to action.

Future Shi – in understanding, trust and autonomy

– In your opinion, what areas of development of AI will become key in the coming years?

-Thee of the main trends is the emergence of highly specialized SHI systems focused on one specific area. These are no longer universal chat bots, but, for example, medical consultants who can accurately interpret the tests and symptoms, or law assistants working in a particular country and in the field of law. Such models will study specialized, carefully tested data and take into account the functions of a professional environment. As a result, 99.9% can achieve the accuracy of their work – and this is the level at which AI can trust serious decisions.

However, such models will become much more compact and cheaper to use. Previously, powerful AI needs servers and clouds, then many of these systems can be launched directly on ordinary devices – laptops, smartphones, even inside corporate applications. This will open the way to the mass and private use of AI without constant dependence on the cloud or the Internet.

The second works with long contexts. Previously, the model could cover a relatively small amount of information. Today she learns to work on goals, correspondence, medical history, research reports. This opens up new opportunities for AI in areas where it is important to consider dozens of interconnected factors.

The third important trend is that AI better understands what they want from. Today, the models are much better to follow the instructions – this is enough to explain once, and AI does exactly what you need.

Why is it important? Because now developers can trust the result without wasting time on constant checks. This opens the way to create more complex multi -storey solutions that work automatically and reliably. Imagine: AI not only answers the question, but sequentially analyzes the situation, draws conclusions, recommendations of forms and begins the necessary actions – all this without the need for a person’s intervention at every step.

Such AI is not just an assistant, but also a full member of the teams and processes – reliable, accurate, adaptive.

– Finally: what would you advise for those who just start their journey in the field of AI?

Do not be afraid to try to experiment. AI is an industry where something new appears every day, and if you want to be on the front edge, it is important to constantly study. Start with the grounds – machine learning, data processing, neural networks. Create your own projects, even small ones, and look for ways to use the theory. And remember: AI is not only a code, but also the art of design systems that can solve real problems.

Nikita is smooth – An example of how an engineer can be not only part of the Shiites revolution, but also form it from the inside. His projects – This is a contribution that already forms the future technology and provides us with access to reasonable and effective solutions.