From Papers to Code: ITMO Introduces a System for Verifying Scientific Connections

ITMO University has introduced PAUK, Russia’s first open tool for building a graph of scientific connections. Developed by the university’s Institute of Artificial Intelligence, the system was presented at an open-source software developers’ conference in Pereslavl-Zalessky. Unlike conventional bibliographic databases, PAUK connects not only papers and their authors but also the software code behind the research, creating a unified, verifiable map of research activity.

Anatomy of a Scientific Network

PAUK’s key innovation is its move away from simple keyword searches toward establishing verifiable semantic relationships. The system draws on the open OpenAlex database, parses PDF documents and scans repositories such as GitHub. Particular attention is paid to verifying authors’ identities: the system uses ORCID identifiers or institutional email addresses to confirm who they are. If the evidence is insufficient, the algorithm does not automatically assign a match; instead, it sends the case for manual review. This addresses a persistent problem in scientific research: distinguishing between people with similar or identical names and reducing the risk of discoveries being attributed to the wrong researcher.

Code as a Full-Fledged Research Output

For the IT industry, PAUK represents a shift in how research contributions are recognized. Scientometrics has traditionally relied on citation indexes. But in the age of machine learning, a study’s real value is often embedded in its software code and datasets. PAUK gives code recognition as an independent research output. This can help universities and companies avoid duplicating development work, identify highly specialized experts for interdisciplinary consortia and assemble teams based on demonstrated skills. The structured knowledge graph could also provide an ideal environment for AI agents, enabling them to automatically identify algorithms for practical problems in medicine and industry. Ultimately, this could help bring new technologies to the public more quickly.

From Text to Graph: The Next Stage

ITMO’s development is the logical next step in the evolution of scientific search. In 2022, the world moved toward open research metadata with the launch of OpenAlex. In 2023 and 2024, Russian researchers and institutions, including Moscow State University, eLibrary and the Russian Academy of Sciences, adopted neural-network-based semantic text analysis and graph models. By 2026, the focus had shifted to mapping structural relationships. PAUK offers more than content analysis: it lets users navigate an ecosystem linking researchers, organizations, papers and code. The system does not invent knowledge graphs from scratch. Its distinctive feature is the combination of publications and code with rigorous verification, setting it apart from probabilistic AI models that can produce hallucinations.

Looking ahead, the system is expected to support languages beyond Russian, allowing universities around the world to use it and build hubs connecting institutions across borders.

Opportunities and Challenges Ahead

It is too early to claim that PAUK is a global leader in scientific knowledge-graph technology: international platforms are also developing graph-based tools. PAUK’s export potential will depend on its ability to adapt to local databases. The main challenge is turning a tool developed for internal use at ITMO into publicly accessible infrastructure. That will require independent accuracy metrics, solutions to software licensing issues and safeguards for data protection. The team will also need to demonstrate the system’s effectiveness on large datasets and add support for foreign languages to reach international markets, where demand for sovereign research infrastructure is only beginning to take shape.

Another potential direction is integration with multi-agent systems such as CoScientist. If PAUK provides a reliable map of expertise and code, while other AI agents use that information to design experiments, the result could be a closed loop for automated science. This could create new opportunities for students and early-career researchers. Rather than spending time on routine literature searches, they could focus on developing ideas, drawing on an existing knowledge infrastructure.

The launch of PAUK marks an important milestone. Russian science is moving beyond simply accumulating publications toward building an intelligent infrastructure in which AI connects ideas, people and working algorithms in a unified network. Knowledge graphs like these could form the foundation of future scientific research.

The main difference between PAUK and conventional automated parsing is that the system does not simply guess at connections. At every stage, it looks for evidence to support them. For example, to establish that different papers refer to the same person, matching names alone is not enough. The system checks the researcher’s ORCID (an international researcher identifier) or institutional email address. This helps prevent confusion with someone else, since a person’s name may appear in different forms in different papers, such as ‘Ivanov Ivan’ and ‘Ivanov I.’ Sometimes the code mentioned in a paper was written by someone other than the paper’s author. That is why the system separately checks whether the researcher actually created the code or merely cites someone else’s tool. If there is not enough evidence, the decision is deferred or sent for manual review. This helps prevent errors

Andrey Getmanov

Junior Researcher at ITMO’s Strong Artificial Intelligence in Industry Research Center