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

In January 2018 Oracle highlighted the ubiquity of AI in enterprise systems, noting that while many front‑end developers use Python or C++ for search, recommendation and ad targeting, most business analysts work primarily with SQL.

In July 2019, Didi’s data science team met Ant Financial engineers and discovered SQLFlow, an open‑source tool that translates SQL programs into Python to invoke databases and AI engines, making end‑to‑end AI accessible to analysts.

The collaboration proceeded in three steps: Didi contributed deep business insights, supplied three high‑value models (a DNN classification model, an interpretable model, and an unsupervised clustering model) to SQLFlow, and joined the open‑source community to co‑develop the platform.

Within a month Didi added DNN, tree‑based (XGBoost) and clustering models, all supported by SHAP visual explanations and integrated with Hive. These models serve diverse Didi scenarios such as ride‑hailing, bike‑sharing, and finance, demonstrating the universal applicability of the AI capabilities.

Didi’s chief data scientist emphasized that AI models must be not only accurate but also explainable, enabling business units to understand why predictions occur and to inform strategy, marketing, and product design.

The partners envision SQLFlow evolving into an open‑source marketplace where thousands of SQL‑defined models form a “model shelf” that any business can consume, turning SQL statements into ready‑to‑use AI services.

Both companies continue to open‑source dozens of projects, with SQLFlow as their flagship joint effort, and invite the community to contribute. Project resources include the website https://sqlflow.org and the GitHub repository https://github.com/sql-machine-learning/sqlflow . Interested users can also run the example car‑price prediction model via Docker: docker run -p 8888:8888sqlflow/sqlflow:didi .