Between September 23 and 25, 2026, the SISY 2026 conference (IEEE 24th International Symposium on Intelligent Systems and Informatics)
brought together global experts to discuss intelligent systems. We are proud to announce that our colleagues Ivan Polášek and Lukáš Radoský, representing Gratex International and Comenius University within the InnovAIte Slovakia project, participated in this event.
They presented two papers contributing to the automation of software development and natural language processing.
Explore the abstracts of their presented papers:
Onboarding to an unfamiliar codebase is hard when documentation is missing, outdated, or written at the wrong level. We describe a prototype that combines deterministic static analysis of Python repositories with LLM agents. After cloning a project, the tool parses the abstract syntax trees, builds a compact code index, and passes that index into agents that produce documentation, dependency and class diagrams, an architecture overview, and an interactive chat over the code.
The contribution is the pipeline itself: structured code facts constrain what the LLM can say, so the generated text stays close to what the repository actually contains. A small evaluation with user tasks (SUS, Raw NASA-TLX) and an author-side precision and recall scoring shows the tool helps with first-day orientation. It has trouble with whole-system architectural questions and with dense diagrams.
Semantic textual similarity (STS) aims to measure the degree of semantic equivalence between two text units. It plays an important role in many natural language processing tasks, including information retrieval, question answering, and text summarization. While neural approaches dominate this area, deterministic light-weight solutions are also sought after in certain situations.
In this paper, we propose a knowledge-based approach for computing semantic similarity between sentences using a conceptual dictionary, with optional inclusion of LLM-based inference of additional tree nodes. The proposed approach was evaluated on Slovak texts, achieving improvement over previously reported results with similar approaches on Slovak texts. We analyze the influence of individual parameters of the algorithm on achieved scores, and propose future research directions.
E-mail: info@innovaite.sk
Phone: +421 (2) 5341 1441
