AI 应用研发工程师 / 2027 届AI APPLICATION ENGINEER / 2027
杨宇轩 / Yuxuan Yang
AI 应用研发工程师AI Application Engineer
我把模型能力变成可测量、可审计、可改进的工程系统。工作聚焦可信 RAG、受控 Agent 与评测基础设施。I build AI applications that can be measured, audited, and improved. My work focuses on citation-grounded RAG, controlled agents, and evaluation infrastructure.
代表 AI 项目Selected AI work
每个案例都从问题、系统决策和失败风险出发，再回到工程控制与可复核证据。Each case begins with the problem, system decision, and failure risk, then returns to engineering controls and reviewable evidence.
Semantic Lighthouse
把 RAG、受控 Agent、权限和审计组织成同一条可验证链路。A testable chain connecting RAG, controlled agents, permissions, and audit evidence.
CodePulse
用冻结实验、隔离运行和重复指标比较 Code Agent。Frozen experiments and repeated-run metrics for comparing code agents.
NeedRadar
从公开技术讨论抽取需求，并把误接收问题拆成可复现的失败模式。A public-discussion requirement pipeline with reproducible failure analysis.
Aisleway
完整体验可以发生，结果不必发生。The experience can be complete. The outcome does not have to happen.
原生 SwiftUI 购物体验。浏览、选择、下单、配送进度和记忆留存在应用内，不发生真实付款或配送。A native SwiftUI shopping experience where browsing, choosing, order flow, delivery-like progress, and memory remain in-app without real payment or shipment.
关于与经历About and proof
我在南京邮电大学学习大数据管理与应用，预计 2027 年 8 月毕业。我更关心 AI 系统如何被验证和约束，而不是堆叠模型或框架名称。I study Big Data Management and Application at Nanjing University of Posts and Telecommunications, graduating in August 2027. I care more about how AI systems are measured and constrained than about stacking model names.
南京邮电大学 · 大数据管理与应用Nanjing University of Posts and Telecommunications · Big Data Management and Application
Alibaba UnifiedModel · 6 public PRs