Advancing engineering research through context-aware and knowledge graph-based retrieval-augmented generation.

Area of Science:

• Engineering
• Artificial Intelligence
• Computer Science

Background:

• Large language models (LLMs) excel at language tasks but struggle with technical accuracy in engineering domains.
• Traditional Retrieval-Augmented Generation (RAG) methods face limitations with isolated information and traversing semantically linked technical data.
• Existing RAG techniques are ill-equipped for the complexities of engineering code, standards, and design documents.

Purpose of the Study:

• Introduce novel, deployable RAG-LLMs tailored for engineering applications using the n8n automation system.
• Enhance factual consistency and technical accuracy of LLM outputs in engineering contexts.
• Address limitations of traditional RAG in handling complex, semantically linked technical information.

Main Methods:

• Developed RAG-LLMs on the n8n automation system for engineering domains.
• Employed a contextual RAG approach to improve relevance by aligning retrieved content with query context.
• Integrated RAG with knowledge graph retrieval for deeper semantic understanding of complex technical information.

Main Results:

• Demonstrated improved technical accuracy and factual consistency of LLM outputs for engineering code, standards, and design documents.
• Contextual RAG approach enhanced relevance of retrieved information.
• Knowledge graph integration facilitated profound semantic understanding in complex engineering domains.

Conclusions:

• The developed RAG-LLMs offer significant improvements for technical accuracy in engineering applications.
• Contextual RAG and knowledge graph integration are effective strategies for enhancing LLM performance in specialized domains.
• The study provides practical insights into deployment, strengths, and weaknesses, with shared workflows for reproducibility.