Ilmu Komputer & AI editorial
SiriusDeliver: Automating Data Warehouse Delivery at Tencent
The core problem
Enterprise data warehouses (DWs) underpin business-critical analytics, yet the delivery of warehouse tasks remains a complicated production process. As the authors (Haining Xie, Xiaokai Zhou, Jiaming Yang, Siqi Shen, Ziwei Wang, Yifeng Zheng, Tengyue Xu, Yipeng Shi, Zefang Zong, Yang Li, Peng Chen, Jie Jiang, Debiao He, Xiao Yan, and Jiawei Jiang) observe, this process spans context retrieval, workflow configuration, code generation, platform submission, and failure diagnosis. Although large language models (LLMs) and coding agents have advanced software development, they are insufficient for production DW delivery. Production delivery demands dependency-aware orchestration, lifecycle-aware artifact control, and continuous adaptation to evolving platform practices—requirements that generic coding agents do not satisfy.
To address this gap, the authors present **SiriusDeliver**, an end-to-end delivery automation agent for production warehouse task submission. The system is evaluated through offline datasets and a large-scale production deployment on Tencent Cloud WeData. Offline experiments on real-world warehouse delivery cases show that SiriusDeliver improves delivery success and
Innovation
The authors evaluated SiriusDeliver through offline experiments and a large-scale production deployment on Tencent Cloud WeData.
**Offline experiments**: On real-world warehouse delivery cases, SiriusDeliver improved delivery success and automation efficiency over representative baselines.
**Production deployment**: During a two-month deployment across 6 business teams and 4 warehouse task types, SiriusDeliver served 3,600 monthly active users and supported 18,240 delivery sessions. It achieved an **87.2% end-to-end success rate** and a **73.5% autonomous submission rate**.
**A/B test**: A one-month A/B test showed that SiriusDeliver reduced median delivery time from **228 to 23 minutes** and engineer effort from **95 to 11 minutes**, while maintaining comparable final delivery success. These results indicate that SiriusDeliver not only automates delivery but also significantly reduces the time and effort required from engineers.
The key performance metrics can be summarized as follows:
| Metric | Value |
|--------|-------|
| Monthly active users | 3,600 |
| Delivery sessions (2 months) | 18,240 |
| End-to-end success rate | 87.2% |
| Autonomous submission rate | 73.5% |
| Med
Why it matters
The results demonstrate that SiriusDeliver effectively addresses the limitations of generic LLM coding agents in production DW delivery. The hierarchical delivery agent enables dependency-aware orchestration, which is essential for complex warehouse tasks. The artifact lifecycle control module ensures that artifacts are verified and revised before and after platform execution, reducing failures. The trace-driven skill evolution mechanism allows the system to continuously adapt to evolving platform practices, maintaining high performance over time.
The deployment on Tencent Cloud WeData across 6 business teams and 4 warehouse task types validates the system's scalability and robustness. The 87.2% end-to-end success rate and 73.5% autonomous submission rate indicate that the system can handle a large volume of delivery sessions with minimal human intervention. The A/B test results show a dramatic reduction in median delivery time (from 228 to 23 minutes) and engineer effort (from 95 to 11 minutes), highlighting the practical impact on engineering productivity.
However, the authors note that the system was evaluated on a specific platform (Tencent Cloud WeData) and may require adaptation for other platforms. Future work could explore cross-platform generalization and further improve the autonomous submission rate. Overall, SiriusDeliver represents a significant step toward end-to-end automation of data warehouse delivery, combining LLM capabilities with domain-specific orchestration and lifecycle control.
Who should read this
Opening member content…