Zlatko Joncev

Introduces a guided search method, Tango*, for solving starting material-constrained synthesis planning problems. It showcases how optimizing a single hyperparameter can lead to improved efficiency and solve rates, and highlights the effectiveness of a computed node cost function in guiding synthesis pathways.

This paper focuses on the development and training of a scientific reasoning model specifically for chemistry, indicating advancements in AI's ability to understand and process complex chemical information. Co-authored with Andres M Bran, Theo A Neukomm, Daniel P Armstrong, and Philippe Schwaller.

Explores methods for extracting structured chemical knowledge from total synthesis documents using LLMs and other computational approaches, contributing to the automation and analysis of chemical data.

Demonstrates that large language models (LLMs) can serve as powerful tools for chemical analysis. When integrated with traditional search algorithms, LLMs enable a new approach to computer-aided synthesis that mirrors human expert thinking, guiding search algorithms toward chemically meaningful solutions for strategy-aware retrosynthetic planning and mechanism elucidation.