Next Generation AI-Enhanced Multi-Modal Database Optimizer

We develop the next generation multi-modal query optimizer with the latest AI techniques. In our recent paper LaPuda: LLM-Enabled Policy-Based Query Optimizer for Multi-Modal Data, a large language model (LLM) based query optimizer is designed for efficient multi-modal, semantic query processing.

Combining powerful LLM and traditional database optimization techniques, the new optimizer achieves significant performance to optimize the multi-modal query plans (with both structured and semantic information, covering multiple data modalities), resulting in 3x query execution speedup for complex multi-modal queries. The design scheme also proves the effectiveness of using LLM in modern query optimizers, especially for complex semantic queries. This project is an important step towards practical design of multi-modal data systems in the LLM era.