<![CDATA[Industrial economic resilience depends on the efficiency of after-sales service provisioning, which is often hindered by semantic ambiguity in customer reports and latency constraints of conventional computing infrastructures. This study examines the integration of a Language Processing Unit (LPU) with a Large Language Model (LLM) under a Deterministic Reasoning Architecture (DRA) framework to address these limitations. Experiments were conducted on a heterogeneous dataset (N = 27.500) consisting of operational service records from PT Rekaindo Global Jasa and a Southeast Asian manufacturing entity over the period 2021–2025. Semantic complexity analysis based on Shannon Entropy indicates that the Repair category exhibits the highest information density (5.2 bits), corresponding to an increased risk of logical failure. Performance benchmarking demonstrates that the proposed LPU-based architecture achieves deterministic inference with a Risk Priority Number (RPN) of 42 significantly lower than stochastic GPU-based baselines (RPN > 120). Predictive integrity evaluation yields an AUC–ROC of 0.988 and an inter-rater agreement of 0.81 (Fleiss Kappa), indicating substantial alignment between automated recommendations and expert assessments. Economic robustness is validated through Monte Carlo simulations, showing a 94.2 % probability of achieving Return on Investment within 20 months, even under high-volatility scenarios. Furthermore, the framework complies with ISO/IEC 42001:2023 and the EU AI Act, achieving a Fairness Ratio above 0.94. Overall, the results demonstrate that the LPU–LLM synergy enables fast, reliable, and responsible generative AI deployment in industrial settings.]]>
Copyrights © 2026
