Semantic Intelligence: Graph RAG-Driven Agents for Time Series Analytics

This work introduces a novel framework that combines LLMs with knowledge graphs to enhance the automation, interpretability, and contextual reasoning in time series analytics. It allows users to interact with time series datasets through intuitive, natural language queries, enabling the extraction of actionable insights from both raw data and semantic context. By leveraging Graph RAG, the framework integrates structured domain knowledge, improves reasoning accuracy, and facilitates context-aware responses. Furthermore, leveraging a knowledge graph built upon a dedicated ontology serving as a memory mechanism enables the incremental accumulation of insights and supports iterative, dynamic exploration of temporal characteristics. This approach simplifies access to advanced analytics for users with diverse expertise levels, allowing for a more comprehensive understanding of time series data. This paper outlines the framework’s architecture, assesses its performance through an evaluation using multiple LLMs, and discusses potential future improvements.