李大鹏
AI Agent Platform Architect · Open-source builder · Former iOS Lead
AI Agent 平台架构师 · 开源产品开发者 · 前 iOS 团队负责人
I build AI-native engineering systems that connect specifications, agent execution, CI/testing, and delivery evidence into one reliable loop.
我构建 AI Native 研发系统，把需求规格、Agent 执行、CI／测试和交付证据连接成可靠闭环。
At HelloTalk, I work on enterprise AI agent platforms and developer productivity. I came to AI engineering from iOS, client architecture, and engineering leadership.
我在 HelloTalk 从事企业 AI Agent 平台与研发效能建设；此前长期负责 iOS、客户端架构和研发团队管理。
My focus is not making agents look autonomous. It is making their work controllable, recoverable, and independently verifiable.
我关注的不是让 Agent 看起来更自主，而是让它的工作可控、可恢复，并且能够被独立验证。
开源项目
A local-first control plane for coding agents across phone, tablet, and desktop, with end-to-end encryption and a zero-knowledge relay.
一个本地优先的 Coding Agent 多端控制工具，覆盖手机、平板和桌面端，采用端到端加密与零知识中继。
A native macOS menu bar app that centralizes approvals across concurrent Claude Code sessions.
一个原生 macOS 菜单栏应用，集中处理多个并行 Claude Code 会话的授权请求。
A quiet, offline audiobook player for audio files you already own, built for iOS and Android.
一个安静、离线的本地有声书播放器，面向用户已有的音频文件，提供 iOS 与 Android 版本。
我在意的事
- Evidence before confidence — agents propose; independent systems verify.
证据先于信心——Agent 提出候选方案，独立系统负责验证。 - Clear trust boundaries — especially when tools touch user code and data.
清晰的信任边界——尤其当工具会接触用户代码和数据时。 - Product restraint — define the real problem, keep the system small, and earn every feature.
产品克制——先定义真实问题，保持系统精简，让每个功能都有存在的理由。
当前技术栈
Go · Swift · Kotlin · TypeScript · Python · Kotlin Multiplatform · MCP · GitLab CI
联系我