The value of a scientific journal lies not only in its individual articles, but in the body of knowledge they collectively create. Although scholarly publishing has become largely digital, its underlying infrastructure remains primarily focused on storing and retrieving individual documents. Most journals provide access to content through PDF files and keyword-based search, while more advanced publishing platforms additionally publish articles in structured XML format to support interoperability and indexing in international bibliographic databases. However, the full potential of this structured content remains largely untapped. Advances in artificial intelligence (AI) now make it possible to leverage the same infrastructure for a new role of journals – not merely as repositories of publications, but as active knowledge systems.

Using the journal Cardiologia Croatica as a case study, this workshop presents the transformation of more than 2,000 peer-reviewed articles published over the past decade into a semantic knowledge system. The textual content of structured XML records was converted into vector embeddings and stored in a vector database, enabling semantic retrieval across the journal’s entire archive based on meaning and conceptual relationships rather than keywords alone. This allows users to identify related publications regardless of differences in terminology, trace the evolution of scientific topics over time, and uncover relationships among authors, concepts, and publications that remain hidden using conventional search methods. The platform combines traditional and AI-enabled approaches to accessing scientific content. In addition to browsing by author, keyword, and publication year, it enables semantic literature search using natural language. There is a conversational AI agent, based on a Retrieval-Augmented Generation (RAG) architecture, which was trained on retrieving relevant sections of the journal’s corpus and exclusively uses journal insights to generate responses. The same infrastructure also supports AI-generated educational flashcards, interactive clinical case presentations, and chronological visualizations of the development of scientific topics and the contributions of individual publications.

Alongside presenting the system architecture and implementation process, the workshop will discuss the applicability of this approach to other scientific journals, institutional repositories, and digital archives. The broader adoption of this model has the potential to transform traditional literature archives from collections of individual publications into dynamic knowledge infrastructures that provide more personalized access to scientific research, education, and scientific knowledge.