Qichang Zheng郑其昌 郑其昌Qichang Zheng
I build the environment AI agents run in — and a platform for developers to orchestrate many agents working together on complex projects. 我为 AI Agent 构建运行环境，也为开发者打造在复杂项目中统筹多个 Agent 协同工作的平台。
AI Agent Platform Lead, based in Shanghai AI Agent 平台负责人，现居上海
About关于
I lead the AI Agent platform at Qiji Zhifeng in Shanghai, as tech lead of a 10-person team. We build North Agent Cloud, a platform for building, deploying, and running AI agents; it runs in production across several deployments. 我在上海奇绩智峰负责 AI Agent 平台，是 10 人团队的技术负责人。我们做 North Agent Cloud——一个构建、部署和运行 AI Agent 的平台，已在多个生产环境上线。
Most of my own engineering goes through coding agents. I built Agent Team to run them in parallel — and to catch the ways long-running agents fail. 我自己的大部分工程工作是交给编程 Agent 完成的。为此我写了 Agent Team，让它们并行工作，并专门治理长程任务里 Agent 的典型失效。
Before that I was an AI Scientist at Devz AI in the US, working on enterprise agents, a general-purpose web agent, and fine-tuned domain LLMs. 在此之前，我在美国 Devz AI 担任 AI Scientist，做企业级 Agent、通用网页智能体和垂类大模型微调。
Work作品
North Agent Cloud
A platform for building, deploying, and running AI agents.构建、部署和运行 AI Agent 的平台。
- Scheduling, sandbox isolation, egress control, credential custody, elastic scaling, end-to-end tracing.调度编排、沙箱隔离、出网管控、凭证托管、弹性伸缩、全链路追踪。
- Runtime and sandbox are decoupled: long, low-load LLM calls and short, high-load tool execution scale separately. Lazy start and automatic reclamation fit about 10x more sandboxes on the same resources.运行时与沙箱分离：长时低负载的模型调用和短时高负载的工具执行各自伸缩；懒启动加自动回收，让同样的资源承载约 10 倍的沙箱。
- The platform’s own built-in features — trajectory analysis, building agents by conversation — are themselves agents running on it.平台内置的轨迹分析、对话式构建 Agent 等功能，本身就是跑在这个平台上的 Agent。
North Sandbox
A multi-tenant sandbox service compatible with the E2B protocol.兼容 E2B 协议的多租户沙箱服务。
- Per-template warm pools bring sandbox acquisition down to milliseconds.按模板预热，沙箱认领降到毫秒级。
- JuiceFS over object storage as the sandbox file system: data persists in real time, and volumes can be remounted elsewhere.用 JuiceFS 连接对象存储作沙箱文件系统：数据实时落盘，存储卷可以灵活切换。
North Firewall
An egress firewall built specifically for agents, which also runs on its own as a general firewall and proxy.专为 Agent 设计的出网防火墙，也能作为通用的防火墙和代理服务独立使用。
- A transparent, chained forward proxy: iptables intercepts egress traffic transparently, and CA injection with MITM decrypts and audits HTTPS requests.透明的链式正向代理：iptables 透明拦截出网流量，CA 注入与 MITM 解析、审计 HTTPS 请求。
- Credential isolation with Vault: credentials are injected as placeholders, so agents only ever see the placeholder and the exposure surface shrinks to a single gateway.凭证隔离 Vault：凭证以占位符注入，Agent 只接触占位符，泄露面收敛到单一网关。
Agent Team
A control plane for running coding agents in parallel.让编程 Agent 并行开发的控制平台。
- 5–10 Claude Code and Codex agents a day, 20+ at peak, each in its own git worktree and Kubernetes namespace, with message routing between them.日常并行 5–10 个 Claude Code / Codex（峰值 20+），各自独立的 worktree 和 K8s 命名空间，Agent 之间由平台路由通讯。
- Minute-level patrol and custom Stop hooks catch the failure modes of long-running work, like an agent that says it will keep going and then stops. Nothing counts as done until a second agent has tried to prove it broken.分钟级巡检和自研 Stop hook 治理长程任务的典型失效——比如说了要继续却停下；每项改动都要经另一个 Agent 端到端对抗验证，证伪不了才算完成。
- 1,400+ commits. My merged pull requests went from about 32 to about 212 a month after it went live.1400+ 次提交；上线后我个人月均合入 PR 从约 32 个增至约 212 个。
Experience经历
- 2026.03 — now2026.03 — 至今 Qiji Zhifeng奇绩智峰 AI Agent Platform LeadAI Agent 平台负责人
- 2024.08 — 2026.03 Devz AI (US)Devz AI（美国） AI Scientist
Earlier: Prudential Financial, Machine Learning Engineer (Capstone), and Shannon Investment, NLP Engineer, 2023–2024. 更早：Prudential Financial 机器学习工程师（Capstone 项目）、Shannon Investment NLP 工程师，2023–2024。
Education教育
- 2022 — 2024 University of Chicago芝加哥大学 MA in Computational Social Science计算社会科学硕士
- 2018 — 2022 Xi’an Jiaotong-Liverpool University西交利物浦大学 BSc in Economics and Finance经济与金融学士
Get in touch联系我
Shanghai, China中国上海, ， local time 当地时间