The proposed PhD project aims to develop the cognitive core of a next-generation sovereign Enterprise Resource Planning (ERP) system by designing an intelligent knowledge agent capable of retrieving, contextualizing, explaining, and securely delivering information to both human users and AI agents. Conducted jointly by LIRIS and Infologic, the project addresses the growing need for trustworthy AI systems that operate entirely on local infrastructures while complying with industrial and regulatory constraints.
The research focuses on several scientific challenges that currently limit the deployment of Large Language Models (LLMs) in industrial ERP environments. These include (1) inte-grating heterogeneous information sources (structured databases, technical documentation, logs, and source code), (2) personalizing responses according to users’ roles and expertise,
(3) ensuring semantic confidentiality, (4) providing explainable AI (XAI), and (5) adapting to the continuous evolution of enterprise knowledge and software versions.
To address these challenges, the project proposes a hybrid neuro-symbolic architecture that combines semantic retrieval techniques with symbolic knowledge representations such as bu-siness ontologies and knowledge graphs. The resulting knowledge agent will serve both as a conversational assistant for ERP users and as a semantic service accessible to other AI agents within a multi-agent architecture.

Expected Outcomes

The project is expected to deliver both scientific and industrial contributions. Scientifically, it will advance research in hybrid information retrieval, neuro-symbolic AI, explainable AI, adaptive user modeling, and trustworthy multi-agent systems. It will also produce new me-thods for semantic access control and context-aware information retrieval in complex enter-prise environments.
From an industrial perspective, the project will provide Infologic with reusable AI technolo-gies that can be integrated into its Copilote ERP platform. These technologies will enable faster and more reliable access to enterprise knowledge, transparent and auditable AI-assisted decision making, improved compliance with the European AI Act, and a fully sovereign de-ployment without relying on external cloud-based AI services.

Tasks and Responsibilities

The doctoral candidate will conduct a comprehensive literature review and analyze the hete-rogeneous data sources available within the ERP ecosystem. They will design, implement, and evaluate a hybrid knowledge agent combining dense retrieval, knowledge graphs, and large language models. The work will also include developing dynamic user models from interaction traces, designing mechanisms for semantic confidentiality and adaptive personalization, and creating explainability frameworks inspired by formal argumentation.
In addition, the candidate will implement software prototypes integrated into the ERP plat-form, conduct experimental evaluations using real industrial data, collaborate closely with both academic and industrial supervisors, publish the research results in leading internatio-nal conferences and journals, and contribute to benchmark creation and technology transfer toward industrial deployment.

Intended Impact    

The project seeks to transform traditional ERP systems into intelligent cognitive assistants capable of providing accurate, personalized, and trustworthy access to enterprise knowledge. By combining hybrid retrieval, semantic privacy, and explainable AI, the developed techno-logies will improve user productivity, reduce the time required to locate critical information, and increase confidence in AI-assisted decision making.
More broadly, the research will strengthen the scientific foundations of trustworthy indus-trial AI while contributing practical solutions for the deployment of sovereign AI systems in highly regulated sectors. The project will reinforce the competitiveness of Infologic’s ERP platform and contribute to the development of transparent, secure, and human-centered AI technologies for enterprise applications.