DeNA 将测试设计工具 DAAQ 推广至全公司，测试工作量减少约 50%
DeNA 用 Milvus 搭的 DAAQ 帮 QA 工程师写测试用例，内部测试设计工作量平均降了 50%，一个项目里 95% 的用例不用改就能用。做测试平台或想给 QA 流程加检索上下文的团队可以看他们的架构拆解。
日本科技公司 DeNA 已将 AI 测试设计工具 DAAQ 从早期试点推广到全公司使用，目标是辅助 QA 工程师完成测试设计工作。官方数据显示，DAAQ 平均减少约 50% 的内部测试设计工作量，其中一个项目中 95% 的生成测试无需修改即可直接使用。
在架构上，DAAQ 把工作流程拆分为三个步骤：从需求文档提取信息、识别需要测试的点、起草测试用例，QA 工程师在每一步都可以审查和编辑输出，从而把精力留给风险分析和边界场景。向量数据库 Milvus 负责检索相关的规格说明、界面设计和历史测试知识，为 LLM 提供上下文。
DeNA 选择 Milvus 的原因包括高性能、计算资源可独立扩展、与 Google Kubernetes Engine（GKE）的兼容性，以及对项目数据隔离和访问控制的支持。完整架构细节发布在 DeNA 官方工程博客上。
Congratulations to DeNA on taking DAAQ from early pilots to a company-wide rollout! 🎉 As the leading tech company in Japan, DeNA built DAAQ to assist QA engineers with test design. We’re happy to see Milvus supporting the team’s work. DAAQ extracts information from specs, identifies what to test, and drafts test cases in separate steps. QA engineers can review and edit the output along the way, leaving more time for risk analysis and edge cases. Milvus retrieves relevant specs, screen designs, and past testing knowledge to give the LLM context. DeNA chose it for its high-performance, independent compute scaling, fit with Google Kubernetes Engine (GKE), and support for project data isolation and access control. DeNA reports that DAAQ reduced internal test design effort by about 50% on average. In one project, 95% of generated tests were judged usable without edits. Thank you to the DeNA Engineering team for building with Milvus and sharing the architecture and lessons behind DAAQ. We’re excited to see what’s next! Read their full engineering write-up (in Japanese): engineering.dena.com/blog/2026/10/d… E #builtwithmilvus s
This report is categorized under AITOP's "Products" desk. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include Milvus, DeNA, DAAQ, RAG.
Source attribution: curated from Milvus (@milvusio).
