CADENZA in Action: Breaking the Monolith with Intent-Dependent Plan Spaces for Semantic Queries

2026-07-01

AI Summary

This work proposes an intent-driven plan space decomposition mechanism to address the limitations of existing semantic query optimizers, which typically treat semantic operators as monolithic models and struggle to balance expressive power with execution efficiency. By decomposing natural language queries into multi-step sub-intents, the approach enables independent selection and parameter tuning of physical operators for each step, facilitating fine-grained trade-offs among result quality, latency, and cost. This paradigm departs from conventional monolithic optimization by integrating intent parsing, multimodal data processing, and interactive plan exploration, thereby achieving both high semantic fidelity and computational efficiency in query execution over multimodal databases.

Abstract

Semantic query processing engines execute semantic operators, whose behavior is specified by natural-language intents, via model inference over multimodal data. Most existing optimizers optimize the operators at the granularity of monolithic implementations -- such as LLMs and embedding models -- forcing a trade-off between expensive model calls and cheaper alternatives that fail to capture intent-dependent semantics. We present CADENZA, a semantic operator optimizer that compiles an intent into decomposed steps, selects concrete physical implementations for each step, and tunes their parameters under user-specified quality-latency-cost preferences. In this demonstration, users interact with CADENZA through a web interface over multimodal databases, exploring how an intent is decomposed into alternative plans, how each plan is optimized, and how different preferences yield different winning plans.

Problem

• semantic query processing
• intent-dependent semantics
• operator optimization
• monolithic implementations
• quality-latency-cost trade-off

Innovation

• semantic query optimization
• intent-dependent planning
• plan decomposition
• multimodal data processing
• quality-latency-cost trade-off