A RAG-based LLM Approach for Data Validation and Harmonization in Ship Design

Abstract

<p>Validating ship design data across systems is challenging due to fragmented information from multiple<br>sources, file types, and formats – from 2D drawings, 3D models, and specifications, often found in<br>unstructured text files. While unified 3D models aim to serve as a single source of truth, ensuring<br>accuracy and consistency across all representations remains a complex task. This paper presents a<br>retrieval-augmented generation (RAG) solution for extracting and comparing design parameters from<br>diverse files and formats. The approach aims to detect inconsistencies between documents and versions,<br>helping designers maintain data integrity and reduce manual effort throughout the ship design process.<br></p>