Abstract : Alert fatigue in cloud operations engineering is a measurable reliability risk: sustained high-volume notification processing degrades engineer response quality, increases Mean Time to Resolve (MTTR), and contributes to on-call attrition among experienced staff. Existing Artificial Intelligence for IT Operations (AIOps) tooling improves detection quality but does not address the notification delivery layer, which is where alert fatigue originates. This paper introduces Human-Centric AIOps (HC-AIOps), a framework that applies a formally specified cognitive load-aware prioritization scoring model to the notification delivery layer of the AIOps stack. The central contribution is a priority scoring function P(e, r, c) that quantifies alert urgency as a weighted composite of event severity, business impact, historical urgency, and engineer role relevance, and a feedback learning model that adapts per-engineer scoring weights from behavioral signals. The research question under evaluation is: can a formally scored, cognitively optimized notification framework reduce extraneous cognitive load proxies, alert volume, duplicate notification rate, and false positive alert rate, while improving critical incident response time, compared to rule-based and static AIOps routing baselines? Experimental evaluation in a simulated cloud-native environment of 24 monitored services, 60 operational days, and six virtual engineer role profiles demonstrates a 34% reduction in total alert delivery, a 27% improvement in critical-alert simulated response time, a reduction in duplicate notification rate from 52% to 11%, and an alert relevance score improvement from 31% to 78%. Priority classification F1-score reached 0.89 for critical-class alerts (AUC-ROC 0.94), compared to 0.71 for the best-performing static AIOps baseline. All comparisons were statistically significant at p < 0.01 (paired Wilcoxon signed-rank test). These results confirm that notification systems formally designed for human cognitive architecture produce measurably better operational outcomes than systems engineered for technical detection coverage alone.