The category AI Alma Mater welcomed educational and research organizations, as well as higher education institutions that promote AI technologies and develop training in the field. AI Talent Hub, a Master’s program by ITMO University and Napoleon IT, became a laureate of the award. The program helps over 750 students acquire AI competencies for various positions, including AI engineer, product engineer, data analyst, and data engineer. Students create their individual learning tracks made of courses by practicing IT experts and one of four major directions: industrial, research, startup, or AI in education. The curriculum comprises over 80 courses and includes project work on cases provided by partner companies: X5 Group, Ozon Bank, MTS, Sberbank AI, Nornickel, and Napoleon IT.

“We are pleased to see that educational projects in AI are recognized, and we value the awards that support it. AI Talent Hub aims to bring together active AI specialists and form a big community that will create major projects and support professionals, bringing the entire field forward,” says Dmitry Botov, a co-founder and head of AI Talent Hub.

AI Brotherhood is a category for communities and teams working in the field of AI. One of the laureates in this category is the project MARS (Multi-Agent Requirements Synthesizer), developed by the AIST Lab of ITMO’s Artificial Intelligence Technologies Faculty. MARS is a model that analyzes AI agent behavior and the distribution of informational responsibility between humans and AI. The project was created by students of the AI Engineering Bachelor’s program Denis Pyshkin, Artem-Zhan Shipitsyn, and Irina Kupriyanova and supervised by Anastasia Filatova.

Artem-Zhan Shipitsyn, Denis Pyshkin, and Irina Kupriyanova. Photo courtesy of Denis Pyshkin

Artem-Zhan Shipitsyn, Denis Pyshkin, and Irina Kupriyanova. Photo courtesy of Denis Pyshkin

“AI systems are gradually transitioning from generating separate answers to independently taking on multistep tasks. The more decisions they make without a human, the more important it is to know when to allow such independence. An agent can complete an assignment without correctly comprehending its separate parts. Such mistakes are particularly important at the early stages: for instance, in software engineering, one incorrectly understood request can affect architecture, code, and tests. This leads to excessive corrections and strain on resources, while the result may still not fit a user’s original idea. Our system makes it possible to look at such mistakes in more detail and understand where the AI has lost a key limitation, made an unfounded conclusion, asked an extra question, or made a decision that needed to be approved by a human,” explains Denis Pyshkin, one of the project’s leaders.

MARS evaluates how a model processes information based on five parameters: Preserve (whether the model preserves the given requirements), Clarify (whether it clarifies the missing parts), Infer (whether it can infer information not given directly but evident from the context), Decide (whether it makes independent technical decisions), and Ignore (whether it ignores random notes). Based on this, the system describes the border of informational responsibility between humans and AI, as well as specifies where the model made a mistake. This analysis makes it possible to see the reason for a malfunction not only based on the final result, but also at the preceding stages, where the error occurs.

The team is continuing to develop the project. The next stage is to create a benchmark that will make it possible to assess how well AI agents handle incomplete and ambiguous user requests. The researchers plan to evaluate not only the final result, but also the decisions the agent makes while carrying out the task.

Photo courtesy of Daria Bugrova

Photo courtesy of Daria Bugrova

RuCode.Awards were held at the RuCode national festival in AI and algorithmic programming. The event brings together school and university students, researchers, lecturers, and developers interested in AI applications in science, education, and technology. The festival is organized by the Moscow Institute of Physics and Technology with coordination by the Ministry of Digital Development, Communications and Mass Media.

RuCode.Awards were established in 2025. For the second installment, the expert jury suggested honoring the best educational and research initiatives, technological solutions, AI popularization projects, as well as organizations that implement AI and train future specialists. In 2026, the award introduced the following categories: AI Mentor, AI Talent, Project of the Future, AI Alma Mater, AI Space, Technology of the Future, Generating the Future, AI Brotherhood, and the special category AI Communicator for authors who promote an interest in AI and bring the technology closer to a wider audience. Applications to the awards and festival were open to school and university students, lecturers, researchers, independent authors, bloggers, experts, educational organizations, and teams that implement projects in the field of AI.