Research Group
Tomáš Javůrek
Research areas: Machine Learning, Automated Reasoning, Misinformation Detection
Position: Research Engineer
Tomas Javurek is a Research Engineer with a strong focus on algorithmic optimization and empirical analysis. While recent trends emphasize agentic frameworks, his work concentrates on improving the efficiency and reliability of existing algorithms in both large language models and classical machine learning systems. He has a particular interest in the internal mechanics of LLMs, going beyond surface-level usage to study their behavior, limitations, and optimization dynamics at a deeper level.
His background in particle physics provides rigorous training in statistical inference, numerical methods, and data-driven analysis, which he applies to systematically evaluate and refine model performance. He focuses on both computational efficiency and output quality, aiming to reduce inference cost while maintaining or improving accuracy. He is especially interested in identifying mathematically grounded optimizations and structural shortcuts within complex learning systems.
Selected achievements
- Participated in Rucio Project (CERN, Atlas – Scientific Data Management)
- DisAI: Fact-Checked Article Retrieval System
- Novel Clusterring Method DelTriC
Selected activities
- Equational Reasoning: Developing methods to enable large language models to generate and verify formal mathematical proofs, with a focus on step-wise correctness and symbolic consistency.
- DelTriC: Improving robustness and reliability of the framework, with emphasis on stability under distributional shifts and adversarial or noisy inputs..
- Long-Context Evaluation: Designing and conducting experiments to assess long-context capabilities of LLMs, including degradation patterns, retrieval fidelity, and reasoning consistency over extended sequences. Testing novel attention mechanisms.
Selected Publications
Selected Student Supervising
- Teacher at High-School