DSL-R1: From SQL to DSL for Training Retrieval Agents across Structured and Unstructured Data with Reinforcement Learning

TL;DR

DSL-R1 introduces a unified framework combining logical reasoning and semantic matching through a novel DSL, enhancing retrieval accuracy across structured and unstructured data using reinforcement learning.

Contribution

It presents a new hybrid retrieval approach that embeds vector primitives in SQL-like operators and optimizes DSL generation via reinforcement learning.

Findings

Achieves +12.3% improvement in Hit@1/3 on industrial email benchmark

Outperforms decoupled baseline retrieval systems

Demonstrates robustness in hybrid retrieval tasks

Abstract

Effective retrieval in complex domains requires bridging the gap between structured metadata and unstructured content. Existing systems typically isolate these capabilities, relying on either symbolic filtering or vector similarity, failing to capture their interplay. In this work, we propose DSL-R1, a unified framework that synergizes logical reasoning with semantic matching via a novel Domain-Specific Language (DSL). By embedding vector primitives within SQL-style operators, our approach leverages the complementary strengths of symbolic precision and semantic coverage. We further introduce a reinforcement learning mechanism where rule-based execution feedback and retrieval quality rewards jointly optimize the DSL generation, balancing structural correctness and semantic alignment. Evaluations on a large-scale industrial email benchmark demonstrate that DSL-R1 achieves a +12.3% improvement in Hit@1/3, consistently outperforming decoupled baselines and establishing a robust paradigm for hybrid retrieval.