The work titled “Knowledge-extractor: a self-evolving scientific framework for hydrogen energy research driven by AI agents”, was published on AI Agent (published on December 15, 2025). Finding new materials is vital for global energy needs. However, traditional methods often require significant time and resources. While general purpose AI models show potential, they frequently lack specific expertise and struggle to keep up with the fast changes in hydrogen energy research.
To address this, a team led by Associate Professor Weijie Yang at North China Electric Power University created Knowledge Extractor. The framework uses a Hybrid Knowledge Integration method with a fine-tuned model as its core. This core works with specialized tools like ArxivAnalyzer, PolicyRetriever, and a WebBrowser. These tools allow the system to gather and check new research papers, patents, and policy updates.
Testing with the HydroBench benchmark showed that fine tuning improved the accuracy of the model, allowing it to perform better than several generalist systems. Case studies showed how the agent handles complete research tasks. Examples included analyzing Solid Oxide Electrolysis Cell technology and looking at supply chain risks for hydrogen storage. By turning raw data from many sources into useful insights, Hydrogen Agent acts as a prototype for future scientific tools in various research areas.

DOI:10.20517/aiagent.2025.04