How Tencent’s DEA‑SQL Revolutionizes Text‑to‑SQL for Intelligent BI

Overview

Tencent PGC Big Data Platform shares its research on Text‑to‑SQL, introducing the DEA‑SQL framework and its deployment in the intelligent BI product OlaChat.

1. OlaChat Background, Challenges and Architecture Evolution

Three core challenges: workflow design, functional extensibility, and practical effectiveness.

Three‑fold solution: a human‑like Agent, a pluggable toolbox, and embedded atomic capabilities.

2. DEA‑SQL Solution

DEA‑SQL is a modular workflow that mimics expert reasoning. It consists of intent recognition, controller, memory, reasoning, voting, and self‑correction modules, and relies on prompt engineering rather than model fine‑tuning.

3. Practical Deployment

The system is realized in multiple product forms: a full‑stack web platform, an embedded Copilot in BI dashboards, a code‑assistant “magic wand”, and a Table‑assistant for Excel‑like analysis.

4. Experiments

Evaluated on Spider and BIRD datasets, DEA‑SQL outperforms state‑of‑the‑art methods. Ablation studies show every module is essential, token cost remains moderate, and the approach works with different base models.

5. Product Capabilities

OlaChat supports natural‑language query‑to‑SQL generation, SQL error correction, automatic chart recommendation, and multi‑step data analysis, providing a secure, self‑evolving solution for enterprise data intelligence.