Enhancing AI Reasoning Through Modular Architectures
Sunjhla Handa, K. Suresh, S.K. Sunori, P. Priyanka, A. Adithya, R. Aishwarya · 2026
This paper introduces a modular AI architecture to improve the quality of reasoning through the synergy of neuro-symbolic machine learning, compositional logic, and adaptive resource management. Unlike classic monolithic paradigms that re- quire full retraining, our model can flexibly accommodate new relation arguments, via plug-and- play reasoning modules which both support sequential and compositional reasoning. It is designed to support life-long learning, explainabili1lity, and energy-efficient task execution architecture. Extensive experiments on real-world benchmarks in various fields have been conducted to show the generalizability, scalability, and interpretability of the proposed model. Combining symbolic and neural processing in a modular approach, the novel system overcomes major limitations of existing AI reasoning models and paves the way for scalable, robust, and explainable intelligent agents in mission-critical scenarios.