Software Architecture for Neurosymbolic Generative AI
Radmila Juric, Eiman Almami, Ibtesam Sayedalameen Almami, Yusuf Ardahan Dogru · 2026
Neurosymbolic AI promises to address the widening gap between logic and predictive inference in AI systems. It is the reaction to the dominance of deep learning neural networks and the lack of reasoning. Neurosymbolic models add human readable reasoning to neural networks and may address the trustworthiness and validation of AI results. This paper promotes a generic software architectural model, with the synergy between neural networks and reasoning with logic, which co-habit within the same software applications. The proposal is illustrated with an example from medical science. It shows the options available when using symbolic computing in current non-symbolic AI.