Developing Neuro-Symbolic Conversational Systems for Reliable Multi-Perspective Production Process Intelligence

 

In this tutorial, we discuss the required methodological and development steps to create a new breed of trusted neuro-symbolic conversational systems. These systems integrate LLMs with reasoning engines and process intelligence tools to support accurate, multi-perspective analysis of production processes. Rather than adopting an "LLM-as-reasoner" approach, we advocate an "LLM-as-formalizer" paradigm, in which the LLM is trained to capture the user’s intent, translate it into machine-readable instructions, orchestrate external reasoning and process intelligence tools, and generate actionable insights to analyze and optimize process performance under different scenarios. This hybrid approach ensures that the flexibility of natural language is backed by the sound guarantees of formal solvers, supporting reliability and usability. Through a real-world production process use case, we demonstrate how Information Systems researchers and practitioners can design and implement such systems effectively.