Mohamed Elsheikh, Nicolai Schoch, Mario Hoernicke, Katharina Stark, Thilo Braun, Sebastian Palacio, Nika Strem
In automation engineering, converting unstructured specifications into machine readable formats remains a key challenge. The “Engineering Data Funnel” (EDF) addresses this by combining neuro-symbolic AI and domain knowledge to process multimodal data. This paper presents an EDF component that extracts structured information from process control narratives (PCNs) using a domain‑knowledge‑augmented, schema‑guided LLM workflow with automated validation, improving reliabilityand accuracy over unconstrained LLM prompting.