问HN：构建个人AI头像。哪些技术是必不可少的？
技术问题：如何可靠地将非结构化的跨渠道数据（聊天记录、通话记录、电子邮件线程）结构化/统一为一个一致的“单一真实来源”，而不失去时间线或说话者意图等上下文信息？场景：能否避免遗漏关键细节（例如，通话中客户的口头预算限制与电子邮件中的书面功能请求）？
技术问题：在个性化（需要访问私人数据）与安全性（防止家庭聊天、工作合同泄露）之间，如何找到一个务实的平衡点？场景：端到端加密加上本地数据处理对于非企业用户来说是否可行，还是会影响可用性？
技术问题：如何使头像的“记忆”变得可操作（不仅仅是存档）——例如，将过去的互动与当前任务关联起来？场景：在我计划晚餐时，能否自动提取6个月前关于朋友过敏的聊天记录，而不是让我手动搜索？
查看原文
Technical: How to reliably structure/unify unstructured cross-channel data (chat texts, call transcripts, email threads) into a consistent "single source of truth"—without losing context like timelines or speaker intent? Scenario: Can it avoid dropping critical details (e.g., a client’s verbal budget limit from a call vs. written feature requests in email)?<p>Technical: What’s a pragmatic way to balance personalization (needing access to private data) with security (preventing leaks of family chats, work contracts)? Scenario: Is end-to-end encryption + local data processing feasible for a non-enterprise user, or will it kill usability?<p>Technical: How to make the avatar’s "memory" actionable (not just a archive)—e.g., linking past interactions to current tasks? Scenario: Can it auto-pull a 6-month-old chat about a friend’s allergy when I’m planning a dinner, instead of me manually searching?