DeepEye-SQL: A Software-Engineering-Inspired Text-to-SQL Framework

TL;DR

Text-to-SQL systems suffer from weak system-level reliability due to brittle, uncoordinated components. DeepEye-SQL treats Text-to-SQL as a software-engineering problem and implements an SDLC-inspired four-phase framework with semantic grounding, N-version SQL generation, deterministic verification via a tool-chain, and confidence-aware release. It achieves state-of-the-art execution accuracy on BIRD-Dev (73.5%) and Spider-Test (89.8%) using ~30B open LLMs without fine-tuning, demonstrating that principled orchestration can surpass larger models or task-specific tuning. The results underscore the practical value of structured, verifiable engineering in deploying reliable Text-to-SQL systems for real-world use.

Abstract

Large language models (LLMs) have advanced Text-to-SQL, yet existing solutions still fall short of system-level reliability. The limitation is not merely in individual modules - e.g., schema linking, reasoning, and verification - but more critically in the lack of structured orchestration that enforces correctness across the entire workflow. This gap motivates a paradigm shift: treating Text-to-SQL not as free-form language generation but as a software-engineering problem that demands structured, verifiable orchestration. We present DeepEye-SQL, a software-engineering-inspired framework that reframes Text-to-SQL as the development of a small software program, executed through a verifiable process guided by the Software Development Life Cycle (SDLC). DeepEye-SQL integrates four synergistic stages: it grounds ambiguous user intent through semantic value retrieval and robust schema linking; enhances fault tolerance with N-version SQL generation using diverse reasoning paradigms; ensures deterministic verification via a tool-chain of unit tests and targeted LLM-guided revision; and introduces confidence-aware selection that clusters execution results to estimate confidence and then takes a high-confidence shortcut or runs unbalanced pairwise adjudication in low-confidence cases, yielding a calibrated, quality-gated output. This SDLC-aligned workflow transforms ad hoc query generation into a disciplined engineering process. Using ~30B open-source LLMs without any fine-tuning, DeepEye-SQL achieves 73.5% execution accuracy on BIRD-Dev and 89.8% on Spider-Test, outperforming state-of-the-art solutions. This highlights that principled orchestration, rather than LLM scaling alone, is key to achieving system-level reliability in Text-to-SQL.

Figures (7)

• Figure 1: Key Idea of DeepEye-SQL
• Figure 2: DeepEye-SQL, a plug-and-play Text-to-SQL framework, consistently surpasses prior SOTA methods using $\sim$30B open-source LLMs without any task-specific fine-tuning.
• Figure 3: DeepEye-SQL Overview.
• Figure 4: A case study from BIRD-Dev (QID: 142) where Value-based Linking recovers a critical column (trans.k_symbol) missed by purely LLM-based linking methods.
• Figure 5: Correctness overlap analysis of three SQL generation methods using Qwen3-Coder-30B-A3B model on BIRD-Dev dataset.