Scholarlink

<p>This paper investigates how an AI agent can be used for automated extraction and integration of unstructured data in an industrial engineer-to-order context. Because a&#xa0;large share of organizational knowledge is contained in freely formatted documents, it has so far been only partially accessible for data-driven decision support. Following a&#xa0;design science research approach, we developed a&#xa0;modular, AI-based prototype in collaboration with an industrial partner, in which the agent acts as middleware: it continuously extracts project-specific information from documents, links it to existing data sets via vector embeddings, and enriches it through interpretations of a&#xa0;large language model. The processed data is integrated into a&#xa0;tool for visual process control and made available to users. An evaluation with real company data shows significant improvements in efficiency, transparency, and traceability of information processing. The paper provides transferable design approaches and identifies research needs regarding multimodality, user acceptance, and sustainable integration.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Dokumente als Wissensquelle: KI-gestützte Integration unstrukturierter Projektdaten im Engineer-to-Order Umfeld

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月1日

查看全文
  • Toni Saupe,
  • Stefan Badura,
  • Jürgen Anke

摘要

This paper investigates how an AI agent can be used for automated extraction and integration of unstructured data in an industrial engineer-to-order context. Because a large share of organizational knowledge is contained in freely formatted documents, it has so far been only partially accessible for data-driven decision support. Following a design science research approach, we developed a modular, AI-based prototype in collaboration with an industrial partner, in which the agent acts as middleware: it continuously extracts project-specific information from documents, links it to existing data sets via vector embeddings, and enriches it through interpretations of a large language model. The processed data is integrated into a tool for visual process control and made available to users. An evaluation with real company data shows significant improvements in efficiency, transparency, and traceability of information processing. The paper provides transferable design approaches and identifies research needs regarding multimodality, user acceptance, and sustainable integration.