Leveraging local LLM for enhanced log analysis: integrating Ollama into electronic logging systems

Modern systems for electronic log keeping and trouble log management generate vast datasets of issues, events, solutions, and discussions. However, extracting actionable insights from this information remains a challenge without advanced analysis tools. This paper introduces an enhancement to two log programs—Electronic Log Keeping (elog) and Trouble Logging (TroubleLog) - used in the RHIC control system at Brookhaven National Laboratory. The enhancement integrates Ollama, an open-source, locally deployed large language model (LLM), to facilitate intelligent log analysis. We present a framework that combines MySQL database queries with Retrieval-Augmented Generation (RAG), enabling users to generate period-based summaries (e.g., daily, weekly) and retrieve topic-specific information- such as issues and solutions - through natural language queries. By indexing log data using vector embeddings and interfacing with Ollama’s API, the system provides accurate, conversational responses while ensuring data privacy by avoiding external data sharing. The paper details implementation aspects, including SQL query optimization and prompt engineering, and evaluates performance using real-world log data sets. Results demonstrate improved usability and significant reductions in manual analysis time. This work the potential of local LLMs in domain-specific log management, offering a scalable and privacy-preserving solution for accelerator control systems.