Unlocking deep eutectic solvent knowledge through a large language model-driven framework and an interactive AI agent

This is the authors' abstract. We don't add key points for this paper.

Artificial intelligence (AI) is playing an important role in advancing green chemical engineering, while the lack of data remains a primary challenge in many fields. Deep eutectic solvents (DESs) are a promising alternative to traditional organic solvents. However, the exploration of new DES formulations has long been constrained by trial-and-error research methods, a preference for familiar formulations, and a lack of easily accessible DES databases. This study proposes a framework driven by large language models (LLMs) for accurately and efficiently extracting data in the DES field, accelerating knowledge discovery. By coordinating LLMs and tools through predefined code paths, we extracted 34,027 data records and 9,215 unique DES formulations from 14,602 research articles, achieving an accuracy of over 90%, thereby creating a comprehensive domain knowledge base. An LLM-driven interactive agent has been deployed on an online platform, further facilitating access to this structured data and enabling researchers to overcome data limitations and accelerate the discovery of new DES formulations. • Developed an LLM-driven framework for automated extraction of DES-related data. • Extracted 34,027 records and 9,215 unique formulations from 14,602 articles with over 90% accuracy. • Integrated AI agents with a graph-based retrieval system for interactive querying. • Built a structured DES knowledge base to accelerate formulation discovery in green chemistry.