Didi and Ant Financial Co‑Develop SQLFlow to Bring AI Capabilities to Data Analysts

In January 2018 Oracle’s official blog published an article titled “It’s Pervasive: AI Is Everywhere”, outlining how AI is expanding within enterprise information systems. While internet giants recruit AI experts to build search, recommendation, and targeted advertising services with Python and C++, most business analysts still work primarily with SQL.

In July 2019, several data scientists from Didi’s Data Science team met engineers from Ant Financial in Beijing. Two months earlier Ant’s AI infrastructure team had open‑sourced SQLFlow, a tool that translates SQL programs into Python, invoking databases and AI engines to enable end‑to‑end AI workflows. Didi’s chief data scientist Xie Liang recognized the project’s potential, and the two parties began a joint development effort.

Within a month Didi contributed three high‑value models to SQLFlow: a deep neural network (DNN) classification/prediction model, an explainable model, and an unsupervised clustering model. These models cover a wide range of scenarios, from ride‑hailing and bike‑sharing to financial services, demonstrating the universal applicability of AI‑enabled SQL.

Didi also added key functionalities to SQLFlow, including integration with XGBoost for tree‑based models, support for unsupervised learning, SHAP‑based visual explanations for both deep learning and tree models, and compatibility with Hive databases.

The collaboration envisions a “SQL garden” ecosystem: a large open‑source repository where each SQL statement represents a ready‑to‑use AI model. This model marketplace would allow business users to simply select a SQL that implements a specific commercial logic, accelerating AI adoption across industries.

Senior data scientist Gao Ziyao emphasized that making AI models accessible to analysts not only improves prediction accuracy but also provides interpretability, which is crucial for business strategy, marketing, and product design.

Interested developers are invited to join the community. The project website is https://sqlflow.org , the source code resides at https://github.com/sql-machine-learning/sqlflow , and the Docker image can be run with the following command:

docker run -p 8888:8888 sqlflow/sqlflow:didi