Trustworthy-by-Design Generative AI Assistants for Industrial Troubleshooting: A Knowledge Graph Grounded Architecture
The essential argument, in the authors’ own words.
BEHIND THE PAPER
Show notes.
Manufacturing organisations increasingly face knowledge drain as specialised technical knowledge remains distributed across experienced staff, design documents, troubleshooting guides and historical correspondence. This paper presents an early industrial case study of a trustworthy-by-design generative AI architecture for troubleshooting in manufacturing, combining: human-validated knowledge graph construction, ontology-guided extraction, graph-based retrieval, multi-agent orchestration, query verification and source-oriented explanation. The paper shows how selected trustworthiness requirements, namely technical robustness and safety, transparency and human agency and oversight, can be operationalised during system design. We discuss the industrial context trustworthiness scope, architectural mechanisms and early implementation observations, including the role of retrieval strategy, document structure and human validation. The paper contributes a practical case experience showing how human-centred oversight, structured grounding and lifecycle-oriented design can support more trustworthy generative AI assistants in industrial settings.
Authors: Rohan Jadhav, Emmanuel Papadakis, George Baryannis
Published in: Proceedings of the Second European Workshop on Trustworthy AI (TRUST-AI 2026)
Publication date: 2026-09-02
Read the paper: https://ceur-ws.org/Vol-4254/paper14.pdf
Source license: Creative Commons Attribution 4.0 International — https://creativecommons.org/licenses/by/4.0/
The authors and publisher do not sponsor or endorse this recording.