EUVIMEDEUVIMED entdecken
Crossref

Knowledge-extractor: a self-evolving scientific framework for hydrogen energy research driven by AI agents

Tongao Yao, Yang Yang, Yujie Yan, Xinyi Ou, Mingyang Li, Chenxi Wang, Wuzhe Li, Chenghao Du, Xuqiang Shao, Zhengyang Gao, Weijie Yang

AI Agent · 15.12.2025 · 1 (1)

DOI: 10.20517/aiagent.2025.04 ↗

Kurz gefasst

Fachartikel in der Zeitschrift AI Agent, online erschienen am 15.12.2025. Verfasst von Tongao Yao und 10 weiteren Autor:innen. Die Arbeit wurde bisher 13-mal zitiert (Crossref) und nennt 32 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The rapid evolution of Artificial intelligence (AI) from passive “knowledge co-pilots” to autonomous “research partners” is initiating a paradigm shift in scientific discovery, a frontier now termed Agentic Science. However, applying general-purpose AI systems to dynamic, vertically integrated domains such as hydrogen energy reveals critical limitations, including a lack of deep domain knowledge, an inability to process real-time information, and insufficient autonomous planning capabilities. To address these challenges, we introduce Knowledge-Extractor, a self-evolving scientific framework for building domain-expert AI agents, which we implement and evaluate in the hydrogen energy domain via an agent named Hydrogen-Agent. The core of our framework is a Hybrid Knowledge Integration strategy, which synergistically combines a domain-fine-tuned large language model (LLM) as its "cognitive core" with a continuously updated, non-parametric knowledge base.This architecture is augmented by an autonomous toolset comprising a PolicyRetriever (for extracting information from policy documents), a WebBrowser (for retrieving online sources), and an ArxivAnalyzer (for analyzing scientific papers from arXiv). We demonstrate that through an autonomous knowledge loop, Hydrogen-Agent overcomes the static knowledge limitations of traditional models. Our experiments validate a “specialization effect” where domain-specific fine-tuning enhances factual accuracy on our HydroBench benchmark, outperforming its base model and powerful generalist LLMs. Furthermore, three case studies illustrates the ability of the agent to autonomously conduct complex, end-to-end research tasks, from multi-source data gathering to the generation of a strategic analysis report. Hydrogen-Agent serves as a robust prototype for future scientific agents, showcasing a viable path toward creating domain-expert AI that can accelerate discovery in critical scientific fields.

Zitieren

APA

Yao, T., Yang, Y., Yan, Y., Ou, X., Li, M., Wang, C., Li, W., Du, C., Shao, X., Gao, Z., & Yang, W. (2025). Knowledge-extractor: a self-evolving scientific framework for hydrogen energy research driven by AI agents. AI Agent, 1(1). https://doi.org/10.20517/aiagent.2025.04

Vancouver

Yao T, Yang Y, Yan Y, Ou X, Li M, Wang C, et al. Knowledge-extractor: a self-evolving scientific framework for hydrogen energy research driven by AI agents. AI Agent. 2025;1(1). doi:10.20517/aiagent.2025.04

BibTeX
@article{Yao2025,
  author = {Yao, Tongao and Yang, Yang and Yan, Yujie and Ou, Xinyi and Li, Mingyang and Wang, Chenxi and Li, Wuzhe and Du, Chenghao and Shao, Xuqiang and Gao, Zhengyang and Yang, Weijie},
  title = {Knowledge-extractor: a self-evolving scientific framework for hydrogen energy research driven by AI agents},
  journal = {AI Agent},
  year = {2025},
  volume = {1},
  number = {1},
  doi = {10.20517/aiagent.2025.04}
}
Kostenlos starten →