📄 中文摘要
对话购物助手（CSA）作为一种具有潜力的智能代理应用，在从原型到生产的过程中面临两个尚未充分探讨的挑战：如何评估多轮交互以及如何优化紧密耦合的多智能体系统。尤其在杂货购物中，用户请求往往不够明确，且高度依赖个人偏好，同时受到预算和库存等因素的制约。研究提出了一种实用的蓝图，用于评估和优化对话购物助手，通过一个生产规模的AI杂货助手进行说明。引入了一个多维度的评估标准，将端到端购物质量分解为结构化的维度，以便更全面地理解和提升购物体验。
构建、评估、优化：多智能体消费者助手的持续改进蓝图
📄 English Summary
Build, Judge, Optimize: A Blueprint for Continuous Improvement of Multi-Agent Consumer Assistants
Conversational shopping assistants (CSAs) represent a compelling application of agentic AI, yet transitioning from prototype to production presents two underexplored challenges: evaluating multi-turn interactions and optimizing tightly coupled multi-agent systems. Grocery shopping exacerbates these challenges, as user requests are often underspecified, highly preference-sensitive, and constrained by factors such as budget and inventory. A practical blueprint for evaluating and optimizing conversational shopping assistants is proposed, illustrated through a production-scale AI grocery assistant. A multi-faceted evaluation rubric is introduced that decomposes end-to-end shopping quality into structured dimensions, facilitating a comprehensive understanding and enhancement of the shopping experience.