事件驱动的AI：使用Kafka和Flink构建研究助手
[
[
The rise of agentic AI has fueled excitement around agents that autonomously perform tasks, make recommendations, and execute complex workflows blending AI with traditional computing. But creating such agents in real-world, product-driven environments presents challenges that go beyond the AI itself.
代理型AI的兴起激发了人们对自主执行任务、提供建议以及将AI与传统计算相结合以执行复杂工作流程的代理的兴趣。但是，在以产品为驱动的真实环境中创建这样的代理面临的挑战超出了AI本身。
Without careful architecture, dependencies between components can create bottlenecks, limit scalability, and complicate maintenance as systems evolve. The solution lies in decoupling workflows, where agents, infrastructure, and other components interact fluidly without rigid dependencies.
如果没有仔细的架构设计，组件之间的依赖关系可能会造成瓶颈，限制可扩展性，并在系统演变时增加维护的复杂性。解决方案在于解耦工作流，使代理、基础设施和其他组件能够流畅地交互，而不需要刚性的依赖关系。
This kind of flexible, scalable integration requires a shared “language” for data exchange — a robust event-driven architecture (EDA) powered by streams of events. By organizing applications around events, agents can operate in a responsive, decoupled system where each part does its job independently. Teams can make technology choices freely, manage scaling needs separately, and maintain clear boundaries between components, allowing for true agility.
这种灵活、可扩展的集成需要一个共享的“语言”来进行数据交换——由事件流驱动的强大事件驱动架构（EDA）。通过围绕事件组织应用程序，代理可以在一个响应、解耦的系统中独立地完成各自的工作。团队可以自由选择技术，分别管理扩展需求，并保持组件之间的明确边界，从而实现真正的敏捷性。
To put these principles to the test, I developed PodPrep AI, an AI-powered research assistant that helps me prepare for podcast interviews on Software Engineering Daily and Software Huddle. In this post, I’ll dive into the design and architecture of PodPrep AI, showing how EDA and real-time data streams power an effective agentic system.
为了测试这些原则，我开发了 PodPrep AI，一个 AI 驱动的研究助手，帮助我为 Software Engineering Daily 和 Software Huddle 上的播客采访做准备。在这篇文章中，我将深入探讨 PodP...