Demonstrated that genomic foundation models behave fundamentally differently from text LLMs via entropy analysis, questioning the value of DNA pretraining; will be presented at ICLR - Learning Meaningful Representations of Life workshop 2026: arXiv
Senior Research Officer
Genome Institute of Singapore, A*STAR
Developed GNNome, a GNN-based de novo genome assembler representing a new paradigm for the field; achieved 20% improvement in NG50 and NGA50 over SOTA on human genome with comparable results across three other species; published in Genome Research 2025.
Contributed to the first de novo assembly of an Indian reference genome, providing a valuable genomic resource for the South Asian population: bioRxiv
Lectured at the ASTAR/NTU AI + Computational Genomics course covering deep learning and genome assembly
Intern (ARAP Scholarship Awardee)
Genome Institute of Singapore, A*STAR
Pioneered neural algorithmic reasoning approaches to de novo genome assembly; presented at NeurIPS - Learning Meets Combinatorial Algorithms workshop 2020: arXiv
Developed a CNN-based classifier for long read type classification, achieving competitive accuracy across multiple classes
Research and Teaching Assistant
Faculty of Electrical Engineering and Computing, University of Zagreb, Croatia
Taught introductory Programming and Data Science courses
Education
PhD in Computer Science
University of Zagreb, Croatia
Developed GNNome – a novel approach to de novo genome assembly based on graph representation learning. Supervised by Mile Šikić and Xavier Bresson. Presented my work at a NeurIPS and ICML workshops and published it in Genome Research.
For my Master’s thesis, I worked on machine learning applications in solid-state and statistical physics. Supervised by Vinko Zlatić and Ivor Lončarić.