LLM-Enhanced Data Management
Xuanhe Zhou, Xinyang Zhao, Guoliang Li
TL;DR
This work addresses the limitations of traditional ML in data management by proposing LLMDB, a framework that combines general LLMs with domain-specific fine-tuned LLMs, an LLM executor agent, and vector databases to enable robust, scalable data-management tasks. By separating offline preparation from online inference, LLMDB embeds domain knowledge to reduce hallucinations, lowers cost through semantic search and caching, and improves accuracy via multi-round inference and pipeline execution. The framework is instantiated across three applications—database diagnosis, data analytics, and query rewrite—highlighting modular data preparation, pipeline generation, and knowledge augmentation as core contributions, along with identified research challenges. The proposed approach has the potential to deliver more reliable NL-based analytics, automated diagnosis, and efficient query optimization in real-world data systems.
Abstract
Machine learning (ML) techniques for optimizing data management problems have been extensively studied and widely deployed in recent five years. However traditional ML methods have limitations on generalizability (adapting to different scenarios) and inference ability (understanding the context). Fortunately, large language models (LLMs) have shown high generalizability and human-competitive abilities in understanding context, which are promising for data management tasks (e.g., database diagnosis, database tuning). However, existing LLMs have several limitations: hallucination, high cost, and low accuracy for complicated tasks. To address these challenges, we design LLMDB, an LLM-enhanced data management paradigm which has generalizability and high inference ability while avoiding hallucination, reducing LLM cost, and achieving high accuracy. LLMDB embeds domain-specific knowledge to avoid hallucination by LLM fine-tuning and prompt engineering. LLMDB reduces the high cost of LLMs by vector databases which provide semantic search and caching abilities. LLMDB improves the task accuracy by LLM agent which provides multiple-round inference and pipeline executions. We showcase three real-world scenarios that LLMDB can well support, including query rewrite, database diagnosis and data analytics. We also summarize the open research challenges of LLMDB.