Design and Evaluation of Enterprise Generative AI Solutions for Systems Engineering Applications

The SCK CEN Institute for Science and Technology (IST) has accumulated a substantial body of knowledge over many years, initially through the MYRRHA project and, more recently, within the framework of the ongoing LEANDREA project. This knowledge is stored across extensive repositories comprising technical documentation, requirements, design decisions, engineering constraints, and other project artefacts managed in dedicated knowledge bases and Polarion environments.

Effectively navigating this vast and continuously growing collection of specialized, multidisciplinary, and heterogeneous information is a significant challenge. As a consequence, locating relevant information and reusing valuable experience acquired during previous project phases can be time-consuming and difficult.

To address this challenge, the Nuclear Technology and Engineering (NTE) group has launched several initiatives aimed at improving knowledge access and supporting engineering activities through advanced digital technologies.

Generative AI for Knowledge Management

A first initiative focuses on the application of Generative Artificial Intelligence (GenAI) to technical knowledge management. The objective is to develop intelligent tools capable of:

• Interacting naturally with users through conversational interfaces;
• Answering complex technical questions based on large and evolving knowledge repositories;
• Retrieving, synthesizing, and contextualizing information from multiple sources;
• Assisting engineers in the preparation of technical reports and documentation on user-defined topics.

To support these goals, a prototype system called HOMER has been developed. HOMER is based on Large Language Models (LLMs) combined with Retrieval-Augmented Generation (RAG) techniques and has been implemented in Python. While the current prototype demonstrates promising capabilities, several aspects remain to be further developed and assessed, including system architecture, retrieval performance, answer quality, grounding effectiveness, operating costs, and scalability.

Engineering Parameter Management

A second initiative, developed within the LEANDREA project, addresses the management of engineering design parameters and requirements. To this end, SCK CEN has developed a prototype application called PDBS (Parameter Database System).

Implemented as a Python Streamlit application, PDBS enables users to:

• Explore and visualize relationships among key engineering parameters;
• Create, modify, and manage design parameters;
• Perform what-if analyses and assess the impact of proposed changes;
• Review, approve, or reject modifications;
• Maintain a complete and traceable history of parameter evolution and design decisions.

The tool is intended to support consistency management, requirements engineering, and informed decision-making throughout the system design process.

Thesis Objectives

The proposed thesis will contribute to the further development and evaluation of these innovative tools through one or more of the following activities:

1. Enhancement and fine-tuning of the HOMER and PDBS platforms, including improvements to usability, architecture, performance, and integration with engineering workflows.

2. Quantitative evaluation of HOMER, including:

   • Retrieval benchmark development and execution;
   • Grounding and answer-quality assessment;
   • Analysis of performance, accuracy, hallucination rates, and operational costs;
   • Comparison of alternative LLMs, embedding models, and RAG architectures.

3. Integration of HOMER and PDBS into a unified Streamlit application, enabling:

   • Natural-language interaction with engineering parameters;
   • AI-assisted interpretation of proposed parameter modifications;
   • Automated impact analysis and explanation of design changes;
   • Enhanced support for systems engineering decision-making.

4. Investigation of the applicability of Generative AI techniques to systems engineering processes, with particular focus on requirements management, design-space exploration, knowledge reuse, and engineering traceability.

Candidate Profile

The ideal candidate should possess:

• A background in nuclear engineering, systems engineering, computer science, or a related discipline;
• An interest in artificial intelligence and digital engineering methodologies;
• Good programming skills in Python;
• Familiarity with distributed software environments and version-control systems;
• Basic knowledge of Large Language Models (LLMs), Retrieval-Augmented Generation (RAG), vector databases, or related Generative AI technologies (considered an asset, but not mandatory);
• Strong analytical skills and an interest in applying AI techniques to real-world engineering challenges.

This thesis offers a unique opportunity to work at the intersection of nuclear engineering, systems engineering, and artificial intelligence, contributing to the development of next-generation tools supporting the design of advanced nuclear systems.