安全型 AI 应用Safety-focused AI application· 2026
EasyPhone AI
面向海外华人家庭的安全手机教练A safety-first phone coach for overseas Chinese families
低风险问题一步一步教，高风险场景先停下来，再把必要上下文交给家人。Guide low-risk problems one step at a time; stop on high-risk signals and hand the right context to family.
案例摘要Case summary
目标用户Audience
需要帮助海外华人中老年家人安全使用手机的家庭成员，以及不熟悉数字设备的老人。Families helping older relatives use smartphones safely, and older users who need a low-friction digital coach.
我的角色Role
负责产品定位、安全边界、分步指导、风险中断和家人求助卡的产品决策与验收；代码由 AI 编程工具完成。Owned the product position, safety boundary, step-by-step guidance, risk interruption, and family help-card decisions; code was produced with AI coding tools.
交付结果Outcome
完成低风险分步指导、高风险中断、家人求助卡和 AI 失败本地兜底的可演示闭环，并部署线上 Demo。Delivered and deployed a demonstrable loop for low-risk guidance, high-risk interruption, family handoff, and local fallback when AI is unavailable.
产品洞察Product insight
对不熟悉智能手机的老人来说，最重要的不是一次得到完整答案，而是在当前步骤获得一个安全、可逆、不会误导的下一步。For an older user who is not comfortable with smartphones, the best answer is not a complete explanation. It is one safe, reversible next step that does not invite a dangerous mistake.
项目简介Overview
EasyPhone AI 面向海外华人家庭中的中老年用户，把常见手机问题拆成可执行的下一步；遇到验证码、银行链接、转账或屏幕共享等风险信号时立即停止，并生成家人求助卡。EasyPhone AI helps older users in overseas Chinese families handle everyday phone problems through one actionable next step. When it detects OTPs, bank links, transfers, or screen-sharing risks, it stops and creates a family help card.
技术栈Tech stack
结果证据Evidence
README 记录 M0–M6 已完成，包含核心领域模型、低风险指导、高风险中断、AI 接入和 Demo 部署。The README records M0–M6 as complete, covering the domain model, low-risk guidance, high-risk interruption, AI integration, and demo deployment.
线上 Demo 与演示视频提供了从输入到指导或求助卡的可复现路径。The live demo and walkthrough video provide a reproducible path from input to guidance or a family help card.
AI 不可用时回退到规则与模板，演示不依赖 API Key。When AI is unavailable, the product falls back to rules and templates, so the demo does not depend on an API key.
项目截图Screenshots
关键决策Key decisions
老人遇到复杂手机问题时，长篇解释会增加认知负担，也可能让用户误操作。Long explanations increase cognitive load and can make an older user take the wrong action.
一次只给一个安全、可逆的下一步，并用服务端状态推进确认完成情况。Give one safe, reversible next step at a time and use server-side state to confirm progress.
AI 可能漏判诈骗话术或把高风险问题降级成普通教程。An AI model might miss scam language or downgrade a high-risk situation into an ordinary tutorial.
风险规则作为不可降级的保险丝，AI 只能升级风险，不能绕过停止策略。Use risk rules as a non-downgradable safety fuse: AI may raise risk, but it cannot bypass the stop policy.
想看完整的代码和实现细节？Want the full code and implementation details?
在 GitHub 上查看View on GitHub