Master's Thesis: Multi-Agent Root Cause Analysis for Industrial Engineering

Job description

Req ID:524032 At Alstom, we understand transport networks and what moves people. From high-speed trains, metros, monorails, and trams, to turnkey systems, services, infrastructure, signalling and digital mobility, we offer our diverse customers the broadest portfolio in the industry. Every day, more than 80 000 colleagues lead the way to greener and smarter mobility worldwide, connecting cities as we reduce carbon and replace cars. We are offering a Master’s thesis opportunity in collaboration with academic and industrial partners contributing to the MONA LISA research initiative, which focuses on monitoring and analytics across the whole lifecycle of cyber-physical systems, from concept and models to deployed systems. The thesis is aimed at students who want to work at the intersection of AI, software engineering, industrial analytics, and trustworthy decision support . The proposed topic centres on AI-agent support for root cause analysis in complex engineering environments. In the documented Alstom / MONA LISA use case, AI is intended to support engineers by combining evidence from issue tickets, logs, product documentation, historical cases, source code, and engineering workflow items in order to shorten analysis time and improve issue handling. The long-term ambition is not to replace engineering expertise, but to provide AI support that helps engineers navigate fragmented information more efficiently and with better traceability. Background Engineering organisations working with safety-critical and high-complexity systems often analyse faults and incidents across several disconnected tools and artefacts. In the Alstom / MONA LISA use case for railway systems, the target workflow includes AI-supported log visualisation , root cause analysis , and solution proposal development . The use-case document explicitly describes how an AI agent can retrieve and combine information from Dr MITRAC , Dimensions , SharePoint , GitLab , and EWM in order to support analysis of reported product issues. This challenge is highly relevant because the available information is distributed over different sources, formats, and abstraction levels. Engineers may need to combine operational log patterns, issue descriptions, historical defects, architecture information, source code references, and workflow items before they can form a reliable hypothesis about the most likely cause of an incident. This makes the problem an ideal candidate for AI-agent orchestration , where specialised components can gather, interpret, and validate evidence from different sources before presenting a structured explanation for expert review. Connection to previous thesis work This vacancy is designed as a direct continuation of previous student work already carried out within the same broader collaboration. One previous thesis, “Multi-Modal Document Context Search with Large Language Models for Manufacturing Industries” , investigated how vector-based RAG , graph-based retrieval , and LightRAG can be used to search across industrial documents containing text, tables, diagrams, and other technical artefacts. That work showed that different retrieval strategies perform differently depending on query type, and its presentation explicitly highlighted agentic search as an important future-research direction. A second recent thesis, “An Exploratory Evaluation of LLM-Based Multi-Agent Systems for Specification Update and Task Decomposition in Railway Software Engineering” , investigated a six-agent architecture implemented as GitHub Copilot custom agents in VSCode . That system explored how modular agents could support engineering workflows and evaluated the outputs using criteria such as completeness , correctness , and traceability . Taken together, these earlier works provide two essential foundations for the proposed thesis. The first foundation is technical: they show how industrial information can be retrieved from complex document ecosystems using AI-based retrieval methods. The second foundation is architectural: they show how specialised AI agents can be organised into a structured workflow for engineering support. This new thesis extends both strands by focusing on how a multi-agent system can support root cause analysis in a real industrial use-case setting. Thesis objective The objective of the thesis is to design, implement, and evaluate a multi-agent or orchestrator–worker AI architecture for industrial root cause analysis . The system may include specialised agents that work with specific evidence types, for example an agent for ticket and defect history, an agent for product documentation, an agent for source-code context, an agent for work-item relations, or an agent for log-based clues. A central orchestration component could then combine their findings into a structured root-cause report or ranked explanation for human review. The thesis should not be seen as a narrow coding exercise. It is expected to contribute both to scientific understanding and to industrial AI practice by exploring questions such as how agent roles should be scoped, how evidence should be passed between agents, how outputs can remain understandable and auditable, and how the quality of the final explanation can be evaluated in an engineering setting. Examples of research questions Possible research questions include the following. How should a multi-agent architecture be designed to support root cause analysis across tickets, logs, documentation, source code, and workflow systems? What is the right division of labour between specialised agents and orchestration logic in an engineering-analysis workflow? How should agent outputs be presented so that they remain useful to engineers without hiding uncertainty, ambiguity, or missing evidence? How can the system be evaluated with respect to correctness , completeness , traceability , and practical usefulness? What role should human-in-the-loop review play in validating AI-generated root-cause hypotheses in a safety-critical engineering context? Possible scope and work packages Depending on the student’s interest and available project access, the thesis may include several of the following activities. Literature review on multi-agent systems, LLM-based reasoning, and AI support for engineering diagnostics. Analysis of the documented MONA LISA / Alstom use case for root cause analysis and issue handling. Design of an agent architecture, including role definitions, orchestration strategy, and evidence flow. Prototype development using AI and retrieval components. Evaluation against selected industrial scenarios or representative cases, using structured criteria such as completeness, correctness, and traceability. Reflection on deployment constraints, safety, explainability, and practical adoption in engineering workflows. Research environment and tools used in related work The thesis will be carried out in a hands-on research setting where prior related work has already explored a range of practical AI-development tools and environments. The previously shared multi-agent thesis explicitly states that the system was implemented as GitHub Copilot custom agents in VSCode , and that this environment was already deployed in the working context used for that study. The same thesis explicitly lists Python , FAISS , LangChain Core tools , and Ollama in its hybrid retrieval pipeline and technology choices. The earlier multi-modal document-retrieval work and presentation material also reference Dataiku , LangChain , LightRAG , and vector-database-based retrieval approaches within the broader experimentation landscape. In practice, this means the thesis can be framed in a tooling environment that is already close to the one explored in related work, rather than requiring the student to start from a completely blank technical base. Expected outcomes The expected outputs of the thesis include: a Master’s thesis report with a clear scientific framing and industrial motivation, a prototype or demonstrator showing an agent-based root-cause-analysis workflow, an evaluation of the system using relevant quality criteria such as completeness, correctness, traceability, and usefulness, and recommendations for how AI-agent support for root cause analysis could be extended within the broader MONA LISA ecosystem. All about you We are looking for a Master’s student with a strong interest in AI, agent systems, retrieval-augmented generation, software engineering automation, or industrial data analysis . A suitable background could be Computer Science , Artificial Intelligence , Software Engineering , Data Science , Machine Learning , or a closely related technical discipline. Useful qualifications include: good programming skills, especially in Python , because earlier related work used Python-based AI pipelines, interest in LLMs, agent architectures, or information-retrieval systems , because these are central to the problem space, curiosity about engineering workflows, traceability, and trustworthy AI , because the target domain requires outputs that can be reviewed and validated by experts, and motivation to work in a setting where academic experimentation and industrial use-case relevance go hand in hand. Prior knowledge of railway systems is not required. More important is the willingness to learn from domain experts and to approach the problem systematically and critically. This thesis sits at a very timely intersection of agent development , industrial AI , and engineering productivity . It addresses a real problem already documented in an ongoing research and innovation context, while also opening up broader scientific questions about how AI agents should search, reason, explain, and collaborate in technical domains. For a student interested in building something both intellectually strong and practically relevant, it offers a rare opportunity to contribute to the next generation of AI-assisted engineering support. Application Please apply with your CV , academic transcript , and a brief motivation statement describing your background and why this topic interests you. This is a Master’s thesis opportunity and is best suited for students who want to combine research depth with hands-on prototyping in AI for industrial systems. You don’t need to be a train enthusiast to thrive with us. We guarantee that when you step onto one of our trains with your friends or family, you’ll be proud. If you’re up for the challenge, we’d love to hear from you! Important to note As a global business, we’re an equal-opportunity employer that celebrates diversity across the 63 countries we operate in. We’re committed to creating an inclusive workplace for everyone.