Optimizing LLM decision-making and time-series analysis using DSPy

Nascent developments in Large Language Models (LLMs) have unlocked new possibilities for decision-making and data analysis across diverse domains, from military logistics to medical triaging. This paper explores the integration of DSPy, an emerging AI framework for automated prompt optimization, into two key areas: logistics and time-series analysis. DSPy is leveraged to improve real-time route optimization, enhancing decision-making under uncertainty in modified instances of the Traveling Salesman Problem (TSP) and Pickup and Delivery Problems (PDP). In time-series analysis, DSPy enhances LLM capabilities for data pre-processing and visualization, overcoming challenges in structured data interpretation. When tested on large-scale Machine Learning as a Service (MLaaS) workload datasets, DSPy-optimized LLMs successfully replicated complex visualizations with significantly improved accuracy over vanilla models. Across both applications, DSPy’s automated reasoning framework improves adaptability and efficiency, offering a unified approach to optimizing decision-making and analytical tasks. These findings underscore the broader potential of DSPy to revolutionize AI-driven optimization across diverse, high-stakes environments. We conclude with a discussion of future work to apply DSPy in healthcare settings for emergency room triage, hospital logistics, and predictive disease monitoring.