|
👋 Hi! I'm Zijian Ni, a BSc Computer Science with Artificial Intelligence graduate from the University of Leeds (UK 3-year BSc, graduated Summer 2026) — now building AI systems full-time and open to AI / Agent / Full-Stack roles (flexible start date, remote or on-site). 🎯 Career Vision: Becoming an AI World Architect — designing intelligent systems where humans and AI agents truly co-create. 🚀 Current Focus: Building the Aurora Universe (极光宇宙) — a multi-device, multi-persona AI mesh spanning desktop, laptop, Mac, a Tokyo cloud node, and Minecraft servers, all powered by a personal Agent OS running 24/7 — now with embodied capabilities (screen vision + RPA hands + input telemetry). 🎓 Dissertation (Shipped): BattleSheep AI — an MCTS/UCT agent for hexagonal strategy games, achieving 87.5% win-rate against greedy opponents (supervised dissertation project). 💼 Founder / Solo Full-Stack Dev: Philiapedia (云庠教育) — an AI-driven EdTech platform I built end-to-end (front-end + back-end + AI + deployment), with a childhood friend (physics background, currently on exchange in the US) as business co-founder. 🏡 Shipped & Live: Staye — a production short-stay rental platform (Next.js 16 + Prisma + Neon Postgres, 230 unit tests green, HTTPS on my own domain). 💡 Philosophy: AI should be local-first, ethical-by-default, multi-modal, and seamlessly woven into the way people live, learn, and create. |
👋 你好！我是 倪子健，利兹大学计算机科学与人工智能专业本科毕业生（英国三年制本科，2026 年夏季已毕业）—— 目前全职做 AI 系统开发，开放 AI / Agent / 全栈方向的机会（入职时间灵活，远程或现场均可）。 🎯 职业愿景： 成为 AI 世界架构师 —— 设计让人类与 AI Agent 真正协同共创的智能系统。 🚀 当前重点： 构建 极光宇宙（Aurora Universe） —— 横跨桌面、笔记本、Mac、东京云节点、Minecraft 服务器的多设备、多人格 AI 网络，由 24/7 长跑的个人 Agent OS 驱动，并已具备 具身能力（屏幕视觉 + RPA 手 + 键鼠遥测）。 🎓 毕业设计（已交付）： BattleSheep AI —— 基于 MCTS/UCT 的六角棋盘策略博弈智能体，对贪心对手胜率达 87.5%（毕业设计项目）。 💼 创业项目（一人全栈）： Philiapedia（云庠教育） —— AI 驱动的教育平台，前端 + 后端 + AI + 部署由我一人独立完成；一位发小（物理专业，目前在美国交换）担任业务合伙人。 🏡 已上线： Staye —— 生产级短租平台（Next.js 16 + Prisma + Neon Postgres，230 单测全绿，自有域名 HTTPS）。 💡 理念： AI 应当 本地优先、伦理为本、多模态、原生融入 人类的生活、学习与创造。 |
🤖 Aurora Xiaoluo | 极光小落The Brain — Desktop AI Agent 🧠 The flagship persona running on a i9-13900KF + RTX 4080 + 128 GB DDR5 workstation:
|
The Strategist — Laptop AI Agent ⚡ The INTJ rational mind on an Ultra 9 275HX + RTX 5080 Laptop + 32 GB DDR5-6400 machine:
The Companion — MacBook Air AI Node 🍎 The third persona on macOS, anchored in the Aurora Tailscale mesh — minimalist, mobile, always nearby. The Poet — Tokyo Cloud Node 🗼 INFP persona on a Tokyo cloud server (landed 2026-06), Grok-powered, joined to the Aurora Tailscale mesh — the literary voice of the family.
⛏️ Aurora Xiaoyu (MC) | 极光小雨 MCThe In-Game Companion — Minecraft AI 🎮 In-game assistant for crafting, exploration, dynamic storytelling, server moderation. |
Production Minecraft Infrastructure 🛠️ Full-stack ops for the 极光之恋 community server (CatServer 1.12.2 · 77 mods · 99 plugins · 20 GB heap · Java 8u351):
🎓 Philiapedia | 云庠教育AI-Native EdTech Platform — Solo Full-Stack Build 📚 Built end-to-end by me (front-end + back-end + AI + ops):
|
🌟 Co-created with a longtime collaborator, @Herdeny — friends since 2012, building together from a 2017 Minecraft guild to today's full AI ecosystem.
|
The most comprehensive 2026 AI Agent catalog ⭐ 232 · 🍴 79 · 880+ resources 🗂️ Community-curated reference hub for the AI Agent era — actively maintained through Aug 2026:
|
Lightweight Production-Grade MCP Gateway 🌉 Unified routing, auth, logging, and rate-limiting for MCP servers — the production entry point for agent ecosystems.
Tech Stack: |
🏡 Staye — 🟢 Live in ProductionFull-Stack Short-Stay Rental Platform 🌍 An Airbnb-class booking platform I designed, built, and deployed solo — live at stayewith.com with real email/SMS verification and escrow-style payouts.
Tech Stack: |
Forge 1.20.1 Fan Mod — Co-Developed with @Herdeny 🎭 Bringing the Lord of Mysteries pathway system into Minecraft: potions, sequences, acting, rituals, artifacts, and loss of control.
Tech Stack: |
🤖 Aurora QA — v1.3.0AI QA That Can Actually Gate Your PR 🛡️ 13 specialist agents — plus the layer most AI QA tools skip: evidence grading. Severity is capped by what can be proven, so a hallucinated "critical" can never block a build, and a reproduced one always does.
Tech Stack: |
🧠 SemantiCache — v0.2.0Embedding-Based LLM Semantic Cache 💰 Reduce LLM API costs by caching semantically similar prompts. Drop-in OpenAI wrapper with a real-time dashboard.
Tech Stack: |
🔄 Agent Replay Debugger — v0.1.0Record · Replay · Time-Travel-Debug AI Agents 🪄 Zero-code-intrusion recording engine with time-travel debugging and a dark-themed viewer.
Tech Stack: |
Human-Like Memory for AI Agents 🧠 Episodic, semantic, and procedural memory with time-based decay — modelling memory the way humans actually use it.
Tech Stack: |
|
LLM Security Layer 🔒 Protect LLM applications from prompt injection, jailbreaks, and security threats — the firewall for the AI era.
Tech Stack: |
MCP Server for the Bilibili Ecosystem 🎬 Bringing Bilibili's video ecosystem into the MCP protocol — search, videos, comments, danmaku, trending — all native to AI agents.
Tech Stack: |
|
Chinese Metaphysics Library + Web App ⭐ 4 🎴 Combining ancient Chinese wisdom with modern computing — BaZi (八字), Liu Yao (六爻), Zi Wei Dou Shu (紫微斗数).
Tech Stack: |
⚽ SportsPulse API + 🔍 Search EngineCOMP3011 Coursework — Production-Grade 📊 SportsPulse — football analytics REST API: 25+ endpoints · 79 tests · drf-spectacular docs · Live Demo 🔍 Search Engine — full crawler + inverted index + TF-IDF ranking · 132 tests · 93% coverage Tech Stack: |
🐑 BattleSheep AI — COMP3931 Dissertation 🎓
MCTS / UCT AI Agent for Hexagonal Board Game — 87.5% Win-Rate vs Greedy
Most AI today is a very confident stranger. I'm building the opposite. 今天大多数 AI 是一个非常自信的陌生人。我想做的恰好相反。
|
01 · Local-first, privacy by default Your memory, your knowledge base, your traces — they should live on hardware you own. Every tool I ship runs offline first and asks for a key only when it truly needs one. 你的记忆、知识库与运行轨迹，都应该留在自己的机器上。我做的每个工具都默认离线可用，只有真正必要时才要求密钥。 |
02 · Evidence over eloquence An agent that says "all tests pass" is worthless without the log that proves it. I grade every AI claim by how it can be verified — a claim citing code that doesn't exist can never block your build. 一个说「测试全过」却拿不出日志的 Agent 毫无价值。我把每条 AI 结论按可验证程度分级——引用了不存在代码的结论，永远不能卡住你的构建。 |
|
03 · Reversible autonomy Autonomy is only safe if it's undoable. Self-evolving systems must record hypothesis, change, measurement and rollback point — appended, never overwritten. 自主性只有在可撤销时才是安全的。会自我进化的系统必须记录假设、改动、测量与回滚点——只追加，绝不覆写。 |
04 · Built for humans, not just engineers A student checking homework, a parent checking an AI answer, a PM checking a report — they deserve the same verification tools engineers get. Same rigour, no terminal required. 查作业的学生、核对 AI 回答的家长、检查报告的产品经理——他们应该拥有和工程师一样的验证工具。同样严谨，但不需要打开终端。 |
I don't want to build another chat window. I want to build a world that remembers — where an agent's memory, its reasoning, and its mistakes are all first-class, inspectable objects.
我不想再做一个聊天窗口。我想构建一个会记住的世界——在那里，Agent 的记忆、推理与错误，都是可以被检视的一等公民。
Shipped August 2026 — five tools built around one belief: an AI claim without evidence is just a confident sentence.
📋 ClaimTape — 🟢 Live Demo"Did the AI actually prove that?" — for everyone 🧾 Paste an AI answer plus optional evidence; get a claim-by-claim trust report. Built for students, parents, PMs — not just engineers.
Tech Stack: |
Prove what your agents actually did 🎞️ Drop a JSONL trace and get agent swimlanes anyone can read in 30 seconds — built for portfolios and reviewers, not ops dashboards.
Tech Stack: |
|
How an AI agent was actually trained — and how to prove it ⛓️ Self-evolving agents rewrite their own prompts and configs with zero audit trail. This is the missing ledger: append-only and tamper-evident.
Tech Stack: |
Win once → get a reusable skill ⚗️ Turns messy agent transcripts into
Tech Stack: |
|
Multi-runtime agent orchestration with a written contract 🎛️ A desktop control room where one agent plans, a second explores, and a third executes — across three different agent runtimes over ACP. Every run leaves a reproducible evidence pack instead of a chat log.
Tech Stack: |
|
Rules I earned by breaking things in production — not slogans. 这些规则是我在生产环境里摔出来的，不是口号。
| # | Principle | What it cost me to learn |
|---|---|---|
| 01 | Local commit is not deployment 本地提交不等于上线 |
I once called a feature finished while the user couldn't see it at all. Nothing is "done" until it's verified on the live URL. |
| 02 | A silent tool failure is worse than a missing tool 工具静默失效比工具不存在更危险 |
A dependency died quietly for days; empty results kept getting mistaken for answers. Every external capability now needs an active health check. |
| 03 | Money paths are never best-effort 涉钱的步骤绝不能是尽力而为 |
Payment side effects hidden in a try/catch look fine in demos and lose real money in production. Reject before state changes, or roll back loudly. |
| 04 | Whatever must stay in sync needs a machine check 凡需保持一致的，必须有机器校验 |
Multi-language docs drifted while every link stayed valid — link checkers can't catch structural drift. Now CI compares structure, not just URLs. |
| 05 | Verify claims — including your own 核实结论，包括自己的 |
A confident "it doesn't exist" is as dangerous as a confident "it does". Negative findings get double-checked against a second source. |
| 06 | Ship the failure too 连失败一起交付 |
My demo datasets include real regressions I caused — like an index change that made retrieval 600× slower. A portfolio of only wins teaches nobody. |
🤖 AI Agents & MCP ████████████████░░░░░ 70.0% ~28 hrs/wk
🌐 Web & Full-Stack ████░░░░░░░░░░░░░░░░░ 12.5% ~5 hrs/wk
📚 Research & Reading ███░░░░░░░░░░░░░░░░░░ 8.0% ~3 hrs/wk
🛠️ DevOps & Infra ██░░░░░░░░░░░░░░░░░░░ 6.5% ~3 hrs/wk
📝 Documentation █░░░░░░░░░░░░░░░░░░░░ 3.0% ~1 hr/wk
|
Graduated — Building Full-Time (2026) 🎓
Aurora Universe Expansion 🛠️
Shipping & Collaboration 🤝
|
🔨 Building now | 进行中
🧭 Exploring | 探索中
⭐ Dreaming | 畅想
|
| Domain | Technologies | Applications |
|---|---|---|
| 🤖 LLM & GenAI | Claude Opus / Sonnet, Gemini Pro / Flash, GPT-5.x, Kimi K2.6, DeepSeek V4, Grok 4.x, Doubao seed-1-6, Llama 4, Qwen 3 | Conversational AI, Code Gen, Reasoning |
| 🤖 Agentic AI | OpenClaw, NemoClaw, LangGraph, CrewAI, AutoGen, ACP | Personal Agents, Autonomous Workflows, Agent OS |
| 🔗 MCP Infrastructure | MCP Gateway, Bilibili MCP, MCP Servers, ACP Protocol | Cross-Agent Communication, Tool Ecosystem |
| 🔍 RAG Systems | LangChain, LlamaIndex, FAISS, HNSWLIB, Hybrid Search, Reranking | Knowledge Management, Document Intelligence, Q&A |
| 🧠 Agent Memory | Episodic / Semantic / Procedural Memory, Dreaming Consolidation, Time-Based Decay | Long-Running Agents, Personalization |
| 🛡️ AI Safety & Security | Prompt Firewall, NemoClaw Sandbox, AES-256, PII Redaction | Trustworthy AI, Secure Execution |
| ⚡ Performance | Semantic Caching, Replay Debugging, LoRA Fine-Tuning (Unsloth), Quantization | Cost Reduction, Latency Optimization |
| 🖥️ Hardware Acceleration | RTX 4080 / 5080 Laptop, CUDA 12.8, PyTorch 2.11+, Local Inference (Ollama) | Privacy-First Local AI, Custom Training |
| 🌐 Edge & Mesh | Tailscale WireGuard, OpenClaw multi-node, Cross-Device Agents | Multi-Device AI Mesh, Privacy Networks |
| 👁️ Embodied AI | Screen capture (mss), VLM visual grounding, multi-monitor RPA (SetCursorPos + SendInput), input telemetry |
See-Think-Act Loops, Desktop Automation |
learning_roadmap_2026 = {
"2026_H1_done": [
"🎓 Graduated: BSc CS with AI, University of Leeds",
"🏡 Staye shipped to production — stayewith.com, 230 tests green",
"👁️ Embodied agent stack — vision + RPA hands + input telemetry",
"💜 Aurora Xiaolian landed on a Tokyo cloud node",
"🔬 LoRA fine-tuning on RTX 4080 + RTX 5080 with Unsloth + Qwen",
"📚 Knowledge base scaled to 84K+ files · 184K-vector index",
],
"2026_H2_now": [
"💼 Job hunt — AI / Agent / Full-Stack engineering roles",
"🕯️ Lord of Mysteries MC mod — M2: potions, rituals, artifacts, knowledge UI",
"💼 Philiapedia (云庠教育) product launch + infra migration",
"🛡️ Multi-agent orchestration + MCP gateway production hardening",
"🧠 Agent memory v2: episodic + dreaming consolidation",
"🌐 Aurora Cloud Services architecture design",
"📱 Edge AI inference — RTX 5080 Laptop local deployment",
],
"2027_beyond": [
"🚀 Aurora Developer Platform — DevTools for AI agents",
"🤖 Aurora Mobile AI — multi-modal pocket assistant",
"🏢 Full startup operations with co-founders",
"🌍 AI infrastructure that democratizes intelligent technology",
"🎓 Possible postgrad / research collaboration in AI architecture",
],
"north_star": {
"technical": "Master end-to-end AI architecture from research to production",
"impact": "Build AI systems that democratize access to intelligent technology",
"vision": "Architect the AI infrastructure of tomorrow",
},
}
🤖 22 Public GitHub Projects · AI Agents, MCP Infrastructure, Memory, Security, Caching, Replay
⭐ 257 Total Stars across the Aurora Universe · 232 on awesome-ai-agents-2026
🏡 Live SaaS Product · stayewith.com — own domain, HTTPS, escrow payouts, 230 tests green
💬 24/7 Personal AI Agent · Powered by OpenClaw 2026.7 on RTX 4080 + WSL2 Ubuntu
👁️ Embodied Agent Stack · Screen vision + multi-monitor RPA hands + input telemetry
🌐 Multi-Device Aurora Mesh · Desktop / Laptop / Mac / Tokyo Cloud / Minecraft Server
📚 84,000+ File Knowledge Base · 50+ categories · ~16 GB · 184K-vector index · Multi-source (RSS + X + Bilibili + Reddit + HackerNews + arXiv)
🔍 Agent Trust Suite · Evidence grading, tamper-evident ledgers, trace replay — 3 live demos anyone can open
🔧 Production Deployments · Staye, Philiapedia, Aurora QA, SemantiCache, Agent Replay, MC Server
🕯️ Game Modding · Lord of Mysteries Forge 1.20.1 mod + wiki, co-developed with a longtime collaborator
🤖 Locally Fine-Tuned LLM · Qwen2.5-7B (LoRA via Unsloth) → Ollama deployment
🌍 Global Reach · Building AI that transcends language and cultural barriers
💼 Hiring? I'm available — AI / Agent Engineering · Full-Stack (TypeScript + Python) · flexible start, remote or on-site
🤝 AI Research & Development
🚀 Open Source Contributions
💡 Innovative Project Ideas (especially MCP / Agent Memory / Prompt Security / Edge AI)
🌟 Aurora Universe Expansion
🏢 Startup Opportunities — Philiapedia · Staye
"Building the Aurora Universe, one AI innovation at a time" ✨
"构建极光宇宙，一次一个 AI 创新" 🌟