Abstract
Flow reactors and microreactors are increasingly important for automated chemical synthesis, offering improved safety, cost-effectiveness, stability, etc. However, the complexity of chemical reactions presents significant challenges for microreactor design. To overcome these obstacles, this study introduces a tool that uses large language models (LLMs) to extract essential flow information from scientific literature. A self-optimizing workflow based on neural network reduces the extraction time from 24 to 16 seconds after analyzing only 10 research papers. The collected data is then processed by ensemble learning models, which achieve F1 scores exceeding 70% in classifying flow patterns. These advancements enable the efficient design of microreactor systems in piperacillin synthesis with low error rates. This research provides synthetic chemists with an easy-to-use LLM tool for flow reaction system design. Furthermore, it introduces a paradigm for applying machine learning methods and LLMs even in fields with relatively limited literature.