AI 公益 / 信息可及性概念原型AI for good / information-accessibility concept prototype· 2026
CyberCareCyberCare
从公益海报开始的信息可及性原型An information-accessibility prototype starting with event posters
让 AI 先整理草稿，再由人对照原图核对时间、地点和报名方式。AI prepares a draft; people compare the time, location, and registration details with the source.
案例摘要Case summary
目标用户Audience
整理社区活动通知的工作人员、志愿者，以及需要更方便获取活动信息的接收者；实际需求与使用效果待验证。Staff and volunteers preparing community notices, and recipients who need easier access to event information; actual demand and outcomes remain unvalidated.
我的角色Role
推进场景选择、核对流程、失败边界和原型验收；实现由 AI 编程工具协助完成。Drove scenario selection, review flows, failure boundaries, and prototype acceptance, with implementation assisted by AI coding tools.
交付结果Outcome
公开网页原型已可访问；真实模型已有有限测试记录，尚未验证目标用户效果、机构采用或整体识别准确率。A public web prototype is available. Real-model tests are limited; target-user outcomes, adoption, and overall extraction accuracy remain unvalidated.
产品洞察Product insight
把图片变成文字，不等于把信息变得可靠。原图没有给出的内容需要保留未知，修改后需要重新检测，最终确认也不能代替真实用户验证。Turning an image into text does not make its information reliable. Missing details must remain unknown, revisions need new checks, and confirmation cannot replace evaluation with real users.
项目简介Overview
CyberCare 从一个赛博磕头的互动创意，转向活动信息的整理与人工核对。原型支持海报四字段草稿、修订、检测、逐项确认和导出，并尝试短音频转写与截图单步指导。CyberCare grew from a playful kowtowing-avatar idea into a poster-review prototype. It supports four-field drafts, revisions, automated checks, item-by-item confirmation, and export, alongside experiments with short-audio transcription and one-step screenshot guidance.
技术栈Tech stack
结果证据Evidence
2026-10-24 核对公开网页与只读健康接口；本轮未上传材料，也未调用模型或虚拟人服务。On October 24, 2026, the public page and read-only health endpoint were checked. No materials were uploaded and no model or avatar sessions were started.
海报流程区分合成样例、人工转录与真实识别；检测针对导出版本，修订会使旧检测和确认失效。The poster flow distinguishes synthetic samples, manual transcription, and real extraction. Checks apply to the export revision; edits invalidate earlier checks and confirmations.
多模态记录保留了合成语音中“示例活动室”被识别为“市立活动室”的失败；成功调用仍需人工校对。Multimodal records retain a synthetic-audio error: ASR changed “示例活动室” to “市立活动室”. Successful requests still require review.
自动检测、测试记录和确认操作不代表 WCAG 认证、读屏验收、真实用户效果或公益影响。Automated checks, test records, and confirmation do not establish WCAG certification, screen-reader acceptance, real-user outcomes, or social impact.
项目截图Screenshots
关键决策Key decisions
AI 容易把模糊或缺失的活动信息整理成看起来完整的答案。AI can turn unclear or missing event information into an answer that looks complete.
只整理活动名称、时间、地点和报名方式，缺失内容保持未知，由人对照原图修订并逐项确认。Extract four fields, preserve unknowns, and require source comparison and item-by-item human confirmation.
有趣的角色可能吸引注意，也可能挡住真正需要完成的任务。An amusing character can attract attention while getting in the way of the task.
角色可以跳过、动作可以关闭，用户直接进入工作台；预设互动不包装成实时 AI 对话。Let people skip the character and disable motion; describe preset interactions accurately.
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