This paper introduces a novel framework that bridges analytics, prediction, explanation, and optimization to support power sector planning and operation. The framework leverages large language models (LLMs) to assist experts in formulating policy recommendations. At its core, the framework integrates four layers: (i) dynamic line rating (DLR) analytics to capture operational benefits; (ii) probabilistic prediction with explainability to reveal safety margins, thresholds, and drivers of line capacity; (iii) optimal power flow modeling to quantify dispatch cost savings and renewable curtailment reduction; and (iv) hybrid prompt generation that fuses analytical insights with relevant prior cases to guide LLMs in producing evidence-based, context-aware, and cross-cutting policy recommendations. Applied to a modified IEEE 30-bus system under a simulated heatwave, the framework reduced both dispatch costs and renewable curtailment by over 16%, while maintaining thermal overload risk below 1%. The LLM layer ensures that complex analytical insights are translated into grounded policy recommendations that mitigate hallucination and bridge disciplinary silos. By coupling quantifiable system gains with policy-oriented interpretability, the framework demonstrates a scalable pathway to enhance power system operation and resilience in more complex grid environments.
