正在寻找黑客松队友 & 暑期实习机会 · 2026 年暑假 Available for hackathons & summer internships · Summer 2026
Nick Qin portrait

用产品直觉与跨界手艺，打造 AI 原生 产品的独立开发者。 Indie builder shipping AI-native products with product instinct & cross-domain craft.

我是秦翰（Nick Qin）——一个从国际贸易转型的全栈 & AI 工程师。我独立设计、开发并上线完整产品。已交付 3 款应用，正在带 1 个 LLM 量化金融研究项目，靠的是一套持续学习的闭环。 I’m Nick Qin (秦翰) — an international trade student turned solo full-stack & AI engineer. I design, build, and ship complete products. Three shipped apps. One quant-finance LLM research lab. Powered by relentless learning loops.

看看作品See the work 联系我Get in touch
所在地Based in
上海Shanghai, CN
就读Studying
上海对外经贸 · 前 15%SUIBE · Top 15%
已上线Shipped
3 款产品3 products live
当前Currently
LLM 研究LLM research
01 / 方法论Operating system

像产品经理一样思考，像分析师一样学习，像 独立开发者 一样交付。 I build like a PM, learn like an analyst, ship like an indie hacker.

我不是那种等着拿到详细需求文档才动手的工程师。国际经贸的背景让我读得懂市场和用户；AI 原生工具链（Claude Code、Cursor、Dify、LangGraph）让我能用几天、而不是几个季度去把判断变成产品。 I’m not the engineer who waits for a spec. My background in international economics taught me to read markets and users; the AI-native toolchain (Claude Code, Cursor, Dify, LangGraph) lets me act on those reads in days, not quarters.

01

产品思维优先Product thinking first

每个项目都从我自己真实感受过的痛点出发。写代码之前，先写清楚用户要完成的任务。架构服务于体验——而不是反过来。 Every project starts with a friction I’ve actually felt. I write the user’s job-to-be-done before I write a line of code. Architecture serves the experience — not the other way around.

02

快速学习能力Rapid knowledge acquisition

从零基础 React 到数月内上线 3 款产品。我把陌生领域当作信息压缩问题：扫读全景、抓出承重概念、做个 toy、再交付一个真东西。 From zero React to three production apps in months. I treat unfamiliar domains as compression problems: skim the field, find the load-bearing concepts, build a toy, ship a real one.

03

AI × 领域杠杆AI × domain leverage

对我来说 LLM 不只是套个聊天框——它是研究工具、协作工程师、也是产品形态。我用它给 46.7 万条金融记录打标，同时写着生产环境的 TypeScript。 LLMs aren’t a chatbot layer for me — they’re a research instrument, a co-engineer, and a product surface. I’ve used them to label 467K financial records and to write production TypeScript at the same time.

我的轨迹My arc
2024 —
国际经济与贸易International Trade
上海对外经贸大学 · 全英文项目SUIBE, full English program
在宏观模型里发现数据的力量Found data's power inside macro models
2025 —
转型：写第一行代码Pivot: first line of code
从零学 React · Vibe 出第一款产品Zero to React · Vibe-coded first app
第一个失败项目教会我的比任何成功都多First failed project taught more than any win
2026 — 当前now
独立开发 + 研究Indie + Research
3 款产品上线 · LLM 量化金融研究3 shipped apps · LLM research
看源码、写模型、持续交付Reading source, training models, shipping
小米 MiMo Orbit 激励计划 — 入选 AI 驱动开发与创新方向 Xiaomi MiMo Orbit Incentive Program — selected for AI-driven development & innovation
关于我About me

我不小心爱上做产品——一旦你体验过把想法变成真实可用的东西，就再也回不去了。 I didn’t choose building products — I tried it once and couldn’t stop. Once you’ve felt what it’s like to turn an idea into something real people use, there’s no going back.

—— 秦翰— Nick Qin

我最初学的是国际贸易。和国际经济、关税政策、贸易融资打了一年半交道之后，我发现真正让我专注的不是宏观的数字和理论，而是亲手去做一些看得见摸得着的东西。 I didn’t start in tech. I spent my first year and a half studying international trade — tariffs, trade finance, macro policy. What I discovered was that the thing holding my attention wasn’t economic models. It was the act of making things.

转折点出现在我把一个粗糙的想法变成第一款上线产品的时候。那种"这个想法是我的，产品也是我的，任何人打开浏览器就能用"的完成感，比任何考试分数都真实。从那时起，我找到了自己最上瘾的事：从零开始理解一个领域、写代码把想法实现、然后把它发布出去。 The inflection point was shipping my first product — rough, real, and live at a URL. The sensation of “this idea was mine, the build was mine, and now anyone with a browser can use it” was more real than any exam score. From that moment I was hooked: understand a domain from scratch, build the thing, and ship it.

我相信最好的学习方式是做出一个真实可用的东西。不是跟着教程临摹，而是面对一个真正的问题、造出解决方案、然后把它放到别人面前接受检验。失败也好，成功也好，每一次交付都让我比上一次更清楚自己在做什么。 I believe the best way to learn anything is to build something real with it. Not a tutorial clone — a real solution to a real problem, put in front of real people. Win or fail, every ship teaches you something that passive learning cannot.

所在地Based in
上海Shanghai
语言Languages
中文 · EnglishChinese · English
咖啡因来源Fueled by
咖啡 + 好奇心Espresso + curiosity
当前沉迷Currently deep into
LLM + 金融交叉LLM × quantitative finance
02 / 精选作品Selected work

三款产品，三种不同的 产品判断。 Three products, three different product bets.

每个产品都源自一个真实的用户摩擦，都给出了有观点的解。先看问题，再看洞察，最后看我是怎么做出来的。 Each one started from a real user friction. Each one ships an opinionated solution. Read the problem, then the insight, then how I built it.

/ 01

DiaryBuddy

AI 第二大脑日记 —— 捕捉碎片，生成意义。 AI second-brain diary — capture fragments, generate meaning.

Next.js 16 Gemini 2.5 Supabase Notion API
在线访问Live → GitHub →
问题Problem

传统日记 App 要求用户「正襟危坐地写」，门槛太高——白天的灵感碎片，到了晚上早就蒸发了。 Traditional journaling apps demand the user “sit down and write properly.” The bar is too high — daytime fragments evaporate before evening.

洞察Insight

把 捕捉 和 整合 分开。白天像聊天一样丢碎片；晚上一键让 Gemini 合成出四段式日记——完整版、要点、导师视角、待办清单。 Separate capture from synthesis. Daytime: drop messy fragments into a chat-like surface. Evening: one tap, Gemini synthesizes a four-part diary — full entry, key points, mentor insight, todo list.

解法Solution

基于 Next.js 16 App Router + Server Actions 构建。Supabase RLS 强制按用户隔离。Notion OAuth 集成，使用 AES-256-GCM 加密存储 token，配合 HMAC 签名 state 防 CSRF。中英文双语界面。 Built on Next.js 16 App Router + Server Actions. Supabase RLS enforces per-user isolation. Notion OAuth integration with AES-256-GCM encrypted token storage and HMAC-signed state for CSRF protection. Bilingual (EN/中文).

/ 02

Brunch & Dinner

跨 Web / iOS / Android 三端的 AI 食谱规划器 —— 一份代码，零渲染分歧。 AI recipe planner across Web, iOS & Android — one codebase, zero divergence.

React Native Kotlin Swift WebView Shell
在线访问Live → GitHub →
问题Problem

跨平台 RN App 在 iOS 与 Android 上的渲染总会有偏差，而且每次 UI 微调都得走完整的 APK / IPA 发版流程。作为独立开发者，我不愿意付这个税。 Cross-platform RN apps suffer rendering drift between iOS and Android, and every UI tweak forces a full APK/IPA release cycle. As a solo dev, that’s a tax I refused to pay.

洞察Insight

如果两端都通过极简的原生壳加载 同一个 Web URL，渲染天然一致，更新也走 Vercel 部署——而不是应用商店。 If both platforms load the same web URL through a minimal native shell, rendering becomes trivially identical and updates ship through Vercel deploys — not app stores.

解法Solution

设计了 WebView Shell 架构：RN Web 构建产物部署到 Vercel，配上骨架级别的原生壳（Android 端单 Activity，iOS 端四个 Swift 文件）。结果：100% UI 一致、下次打开 App 即热更新、原生壳薄到可以手工审计。 Designed a WebView Shell architecture: one RN Web build deployed to Vercel, plus skeleton native shells (single Android Activity, four Swift files). Result: 100% UI parity, hot updates on next app launch, native shells small enough to audit by hand.

/ 03

Genote  生思笔记

面向中国考研 / 考公学生的 AI 学习助手，把开源 RAG 引擎做成产品。 AI study assistant productized for China’s grad-exam students.

Next.js 16 FastAPI LangGraph SurrealDB DeepSeek V3
GitHub →
问题Problem

考研 / 考公学生淹没在资料里。通用笔记软件没有 AI 检索，也没有备考真正需要的间隔重复闭环。 Chinese 考研/考公 students drown in materials. Generic notebooks lack the AI retrieval and spaced-repetition loop that exam prep actually needs.

洞察Insight

不要重造发动机，要重造 产品外壳。基于成熟的开源 RAG 项目（Open Notebook，23k stars），套上备考专用的外壳：倒计时驾驶舱、连击奖励、专注模式、闪卡。 Don’t reinvent the engine — reinvent the product wrapper. Take a proven open-source RAG foundation (Open Notebook, 23k stars) and wrap it in exam-specific affordances: countdown cockpit, streak rewards, focus modes, flashcards.

解法Solution

FastAPI + LangGraph 管线，接入 18+ 模型供应商。基于 SM-2 算法 的间隔重复，配合翻卡动画。三档专注模式（⌘. 切换）。视频转写文本 ↔ AI 引用的时间戳跳转。oklch 配色的暖琥珀深色主题。Docker Compose 一键部署，定价目标 ¥15–25 / 月。 FastAPI + LangGraph pipeline backing 18+ model providers. SM-2 spaced repetition with flip-card animations. Three-tier focus mode (⌘. toggle). Video transcript ↔ AI citation timestamp jumps. Warm-amber dark theme in oklch. Docker Compose one-shot deploy, designed for ¥15–25/month pricing.

03 / 研究Research

把 LLM 当作 研究仪器，而不是聊天机器人。 Using LLMs as research instruments, not chatbots.

在上海对外经贸大学，我在冯小兵教授指导下主导一个 LLM 研究项目——识别国内投资者互动平台上的 AI 生成恶意机器人行为。 At SUIBE I lead an LLM-driven research project under Prof. Feng Xiaobing — detecting AI-generated malicious bot behavior in China’s investor relations platforms.

金融问答中的 LLM 恶意机器人识别 LLM-Based Malicious Bot Detection in Financial Q&A

2026.01 — 至今present

在深交所的「互动易」平台，上市公司必须公开回复散户投资者的问题。我们发现其中相当一部分回复具有 恶意机器人特征——模板规避、情绪操纵、选择性披露。我设计了三层识别架构：规则筛选 → DeepSeek R1 作为语义裁判 + 五维评分量表 → 置信度聚合。同时还搭建了跨模态对齐管线，用 jieba + TF-IDF + 余弦相似度把业绩电话会议的音频转写与文本问答对齐；并基于 Claude 构建了 Obsidian 知识 wiki（50+ 互链页面），成为团队不断生长的研究记忆。 On the Shenzhen Stock Exchange’s 互动易 platform, listed companies must publicly answer retail investors. We found a meaningful share of those responses exhibit malicious bot characteristics — template evasion, emotional manipulation, selective disclosure. I designed a three-layer detector: rule-based filtering → DeepSeek R1 as semantic judge with a 5-dimension scoring rubric → conviction-level aggregation. I also built the cross-modal pipeline aligning earnings-call audio transcripts to text Q&A using jieba + TF-IDF + cosine similarity, plus a Claude-powered Obsidian knowledge wiki (50+ interlinked pages) that serves as the team’s living research memory.

467,196
问答记录分析Q&A records analyzed
1,355
LLM 全量打分样本LLM-scored full sample
5-维dim
恶意度评分量表Malice scoring rubric
< ¥50
LLM API 总成本Total LLM API cost
04 / 技术栈Stack

真正 用来交付 的工具。 Tools I’ve shipped with.

这不是教程清单——下面每一项都出现在已上线的产品里，或者在使用中的研究产物里。 Not a checklist of tutorials — every tool below appears in a product that’s live or a research artifact that’s in use.

前端Frontend
React Next.js 16 React Native TypeScript Tailwind CSS shadcn/ui
后端 & 数据Backend & data
FastAPI Python Supabase (RLS) PostgreSQL SurrealDB LangGraph
AI / 大模型AI / LLM
DeepSeek R1 / V3 Gemini 2.5 OpenAI API Dify Workflows 提示词工程Prompt Engineering Whisper ASR
构建 & 交付Build & ship
Claude Code Cursor Vercel Docker OAuth 2.0 AES-256-GCM
05 / 我的学习方式How I learn

我真正的优势不是任何一项技能，而是这个 闭环。 My unfair advantage isn’t any single skill — it’s the loop.

一套从「我不懂这个」到「我把它做出来了」的可复制系统。 A repeatable system for going from “I don’t know this” to “I shipped this.”

先压缩，再动手Compress, then build

面对一个陌生领域，我先花 2–3 小时通读核心资料，写一页心智模型，然后立刻做一个最小的真实小项目。Toy 就是检验。 For any new domain I skim the canonical sources for 2–3 hours, write a one-page mental model, then immediately build the smallest real thing that uses it. The toy is the test.

AI 是队友，不是拐杖AI as a teammate, not a crutch

Claude Code 和 Cursor 是结对编程的搭档，Dify / LangGraph 是研究助理。底线是：合并之前永远先理解 diff——AI 负责加速，我负责架构。 I use Claude Code & Cursor as pair programmers and Dify/LangGraph as a research assistant. The discipline is to always understand the diff before I accept it — the AI accelerates, but I architect.

每次都读源码Read the source, every time

跟教程学很快遇到天花板。做 Genote 之前，我把 23k star 的 Open Notebook 源码从头到尾读了一遍；做 WebView Shell 时读了 RN 的桥接代码。代码是最诚实的文档。 Tutorial-driven development hits a ceiling fast. For Genote I read 23k-star Open Notebook source end to end before forking. For the WebView Shell I read the RN bridge code. Code is the most honest documentation.

公开交付，持续迭代Ship publicly, iterate fast

每个项目都挂着我的真实 URL 上线。Commit 历史公开、部署公开、学习过程公开。作品可达，口碑才会复利。 Every project goes live behind a real URL with my name on it. Public commit history, public deploys, public learning. Reputation compounds when work is reachable.

代码之外Beyond code

写代码占我一半的清醒时间；另一半也很值得说。 Code takes half my waking hours. The other half is worth mentioning too.

几个让我保持平衡、补充能量、持续迭代的东西。 A few things that keep me balanced, energized, and iterating.

古典音乐Classical music

交响乐和歌剧是我深度工作的配乐。马勒和德沃夏克是调试代码时的首选——交响乐的复杂肌理和架构和写前端出奇地像。 Symphonies and opera are my deep-work soundtrack. Mahler and Dvořák for debugging sessions — the layered complexity of an orchestra mirrors the structure of a well-architected frontend.

阅读与持续学习Reading & learning

读的内容很杂：技术文档、社科、传记、偶尔读财报。对新领域的基本好奇心是我最大的竞争优势——如果有什么值得学，我就没理由不去弄懂它。 My reading is broad: technical docs, social science, biographies, occasional earnings reports. Raw curiosity about new domains is my biggest edge — if something is worth understanding, I have no excuse not to learn it.

中文 · 英文Chinese · English

中文母语，英语流利——读源码、写文档、和老外组队打黑客松都没问题。双语不只是沟通工具，它让我能用两种文化的视角去看同一个产品问题。 Native Chinese, fluent English — comfortable reading source code, writing docs, and hacking in international teams. Bilingual isn't just communication; it's two cultural lenses for the same product problem.

当前副业Side quests

探索 LLM 如何理解金融市场情绪——不只是因为好玩，而是想让自己的技术真正理解一个学科。最近还在研究端侧 AI 和高性能 React Native 架构，因为手机上的每一毫秒都很重要。 Exploring how LLMs can understand financial market sentiment — not just for fun, but to see if my tech can genuinely comprehend a discipline. Also diving into on-device AI and performant React Native architecture, because every millisecond matters on mobile.