RedParrot: Accelerating NL-to-DSL for Business Analytics via Query Semantic Caching

Researchers at Xiaohongshu have developed RedParrot, a framework that accelerates the conversion of natural language queries into domain-specific languages (DSLs) for business analytics. The system uses a semantic cache to bypass costly multi-stage LLM pipelines by matching new requests against cached query skeletons—normalized structural patterns—and adapting their corresponding DSLs. This addresses the high latency, cost, and error propagation issues in existing NL-to-DSL systems, which are unsuitable for enterprise-scale deployment. The core technical contributions include offline skeleton extraction, online skeleton matching, and DSL adaptation. The work is published on arXiv.