Initiating coverage
首次覆盖
Chaoran Jiang
蒋超然
Data Science, University of Wisconsin-Madison. Equity Research Intern, Greenwoods Asset Management, summer 2026.
威斯康星大学麦迪逊分校，数据科学专业。景林资产股票研究实习生，2026 年夏。
- Ticker
- CJ
- Sector
- Human capital
- Coverage initiated
- 21 Aug 2026
- Analyst
- The subject
- Source notes
- 16, all clickable
- 代码
- CJ
- 所属板块
- 人力资本
- 首次覆盖日期
- 2026-08-21
- 分析师
- 标的本人
- 溯源脚注
- 16 条，均可点击
Three reasons to take the interview.
建议面谈的三条理由
Every claim below carries a bracketed note. Click one and it opens the experience it came from. That is the whole argument, made twice: once in words and once in how the page is built.
下面每一处论断都挂着方括号里的编号，点开就是它背后的那段经历。这既是论证，也是论证方式：说出口的每个数字都能追回源头。
A combination that does not usually come bundled.
这三样东西通常不长在同一个人身上
Buy-side research training, founder-side execution and hands-on AI engineering normally show up in three different candidates. This summer they ran in parallel in one person: a coverage workload on the Greenwoods desk in Shanghai, and a product he owns end to end being built alongside it.
买方研究训练、创业者的执行力、亲手写 AI 工程，通常分散在三个候选人身上。今年夏天它们在同一个人身上并行：一边在上海景林的研究台上跑覆盖，一边推进自己从头负责到尾的产品。
- Desk output. A global AI-video industry market-sizing report produced under a buy-side mentor, delivered as a clean front-facing report over a fully traceable Excel backend.
- Coverage breadth. AI-application equities across US, Hong Kong and A-share markets, synthesizing Goldman Sachs and CITIC-tier sell-side research into a 33-ticker sector dashboard running on a market-data pipeline he built himself.
- Founder side. Ignite 3D Hologram, started at fifteen: a hologram device, a national utility-model patent, a design patent, and two venture pitches that each sent him back to revise the plan.
- 研究台产出。在买方 mentor 指导下完成全球 AI 视频行业市场规模测算报告，前台报告干净，后台 Excel 每个数字都能点回来源。
- 覆盖广度。覆盖美股、港股、A 股的 AI 应用类公司，综合高盛、中信级别的卖方研究，并在自建的行情数据管线上搭起 33 只标的的板块看板。
- 创业一侧。15 岁创办 Ignite 3D 全息：做出全息设备，拿到一项国家实用新型专利和一项外观设计专利，两次向风险投资人路演，每次回来都改一版方案。
The method is the asset. Every number links back to where it came from.
方法本身就是资产：每个数字都追得回源头
Market bases get triangulated across sources and restated to one global scope so they are actually comparable. Penetration is written down as an explicit forward-looking assumption instead of being smuggled into the base. The front page stays clean, the backend stays traceable, and any figure can be walked back to its source in one step. Junior analysts usually take years to build that habit. He arrived at it in a summer because he was made to.
市场基数要多来源交叉验证，并统一折算到同一个全球口径上，彼此才可比；渗透率写成显式的前瞻假设，而不是悄悄揉进基数里。前台报告保持干净，后台保持可溯源，任何一个数字一步就能查到出处。这套习惯多数初级分析师要练几年，他在一个夏天里练出来了，因为研究台上就是这么要求的。
- Sizing discipline. Multi-source triangulation, unified global scope, penetration stated as an assumption rather than an estimate.
- Process, not instinct. A 7-step investment decision framework co-developed on the desk: market selection, a thesis carrying its own falsification signals, staged position sizing. Reviewed by a senior buy-side investor.
- Proof of habit. This page runs the same rule. Sixteen source notes, each clickable, each pointing at the record behind the claim. Nothing here is unsourced.
- 测算纪律。多来源交叉验证，统一全球口径，渗透率作为显式假设单独列出，不当成既有数据用。
- 靠流程，不靠直觉。在研究台上共同搭建 7 步投资决策框架：选市场、写出带可证伪信号的投资命题、分步建仓，并由资深买方投资人复核。
- 习惯的证据。这一页用的是同一套规矩。16 条溯源脚注，条条可点，条条指向背后的经历，没有一句话是没出处的。
AI leverage with the judgment kept human.
用 AI 放大产能，判断权留在自己手上
He runs multi-agent collection, cross-checking and scraping in parallel, then owns every assumption personally. The systems he builds are held to the same standard: whatever comes out has to be explainable to the person reading it, or it does not ship.
信息收集、交叉核对、数据抓取交给多智能体并行跑，假设和判断由他自己认领。他做的系统守同一条线：输出必须能对读的人解释清楚，否则不发。
- Research workflow. AI-assisted information collection, cross-checking and data scraping, with every judgment and assumption owned by him.
- Applied to markets. A daily-frequency US equity system: an interpretable selection model with logistic regression on price, volatility, drawdown and volume features, picking 3 to 5 names out of a roughly 60-ticker liquid universe under integer-share, cash-buffer and sizing constraints.
- Controls before conviction. Data validation and freshness checks against look-ahead bias, evaluated in paper trading with drawdown-focused risk control. LLMs translate the constraints into rule-based logic, which keeps the whole thing readable.
- 研究流程。AI 辅助的信息收集、交叉核对与数据抓取，每一个判断和假设由他本人负责。
- 用到市场上。一套美股日频系统：基于价格、波动、回撤、成交量特征的可解释选股模型，用 logistic regression 估计次日涨跌概率，在约 60 只高流动性标的池中选出 3 到 5 只集中持仓，并受整数股、现金缓冲、仓位规则约束。
- 先有风控，再谈信心。数据校验与新鲜度检查防止用到未来数据，以 paper trading 评估，风控盯回撤。投资约束由 LLM 翻译成规则逻辑，整套系统保持可读。
Exhibit 1Key data
| Item | Value | Note |
|---|---|---|
| Name / ticker | Chaoran Jiang (CJ) | |
| Degree | B.A. Data Science, University of Wisconsin-Madison | |
| Enrolled | Sep 2023 to Jun 2027 (expected) | |
| Buy-side desk experience | Greenwoods Asset Management, Shanghai Jun-Aug 2026 | |
| Sector dashboard built | 33 tickers, self-built market-data pipeline | |
| Markets covered | US, Hong Kong, A-share | |
| Decision framework | 7 steps, reviewed by a senior buy-side investor | |
| Portfolio construction | 3 to 5 names from a ~60-ticker universe | |
| Data sources unified | 3 heterogeneous sources into one model (Course Planner) | |
| Product output | Three or fewer ranked, conflict-free schedules per degree audit | |
| Patents | 2 (national utility model, design) | |
| Prior analytics internship | xMEMS, Santa Clara CA May-Aug 2024 | |
| Toolchain | Python, SQL, Tableau, advanced Excel, multi-agent orchestration | |
| Work authorization | Authorized to work in the U.S. without sponsorship |
Source: Appendix B, notes 1 to 16. No estimate, no valuation and no forecast appears anywhere in this table.
图表 1关键数据
| 项目 | 数值 | 脚注 |
|---|---|---|
| 姓名 / 代码 | 蒋超然 (CJ) | |
| 学历 | 威斯康星大学麦迪逊分校，数据科学学士（B.A.） | |
| 在读时间 | 2023-09 至 2027-06（预计） | |
| 买方研究台经历 | 景林资产（上海） 2026-06 至 2026-08 | |
| 板块看板 | 33 只标的，自建行情数据管线 | |
| 覆盖市场 | 美股、港股、A 股 | |
| 决策框架 | 7 步，经资深买方投资人复核 | |
| 组合构建 | 约 60 只标的池中选 3 到 5 只 | |
| 数据源整合 | 3 个异构数据源并为一套模型（选课助手） | |
| 产品输出 | 每份学位审计生成不超过 3 套无冲突课表 | |
| 专利 | 2 项（国家实用新型、外观设计） | |
| 此前分析实习 | xMEMS（美国圣克拉拉） 2024-05 至 2024-08 | |
| 工具栈 | Python、SQL、Tableau、高级 Excel、多智能体编排 |
资料来源：附录二脚注 1 至 16。本表不含任何估值、盈利预测或目标价。
Exhibit 2Roles and builds, most recent first
| Period | Role | Organization | What was delivered | Note |
|---|---|---|---|---|
| Jun-Aug 2026 | Equity Research Intern | Greenwoods Asset Management, Shanghai | Market-sizing report with traceable backend, 33-ticker dashboard, 7-step decision framework | |
| In development | Founder & Product Owner | UW-Madison Course Planner | Degree audit to ranked conflict-free schedules; Python/FastAPI plus a constraint engine | |
| Independent | Builder | AI-assisted quantitative trading system | Daily US equity pipeline, interpretable selection model, paper trading with drawdown control | |
| May-Aug 2024 | Data Analyst Intern | xMEMS, Santa Clara, CA | Finance data cleaned in Python into the accounting system, Excel reconciliation, Tableau dashboard | |
| Aug 2020 to Jun 2023 | Founder | Ignite 3D Hologram Co. | 3D hologram device, national utility-model and design patents, two venture pitches |
Source: Appendix B. Course Planner is in development; no launch metrics are claimed because there are none yet.
图表 2经历与项目，按时间倒序
| 时间 | 角色 | 机构 / 项目 | 交付成果 | 脚注 |
|---|---|---|---|---|
| 2026-06 至 08 | 股票研究实习生 | 景林资产（上海） | 带可溯源后台的市场规模测算报告、33 只标的板块看板、7 步投资决策框架 | |
| 开发中 | 创始人兼产品负责人 | UW-Madison 选课助手 | 学位审计转成排序过的无冲突课表；Python/FastAPI 后端加约束引擎 | |
| 独立项目 | 开发者 | AI 辅助量化交易系统 | 美股日频流水线、可解释选股模型、paper trading 与回撤风控 | |
| 2024-05 至 08 | 数据分析实习生 | xMEMS（美国圣克拉拉） | Python 清洗财务数据入账务系统、Excel 对账、Tableau 看板 | |
| 2020-08 至 2023-06 | 创始人 | Ignite 3D 全息 | 3D 全息设备、国家实用新型与外观设计专利、两次风投路演 |
资料来源：附录二。选课助手仍在开发中，尚未上线，因此不列任何运营数据。
How a number gets into a report.
一个数字是怎么进到报告里的
This is the process behind the AI-video market-sizing work, and it is the reason the rest of this page is built the way it is.
这是 AI 视频行业测算背后的流程，也是这一页为什么这么搭的原因。
Exhibit 3Front-to-back research workflow
Collect
Market bases pulled from multiple sources, sell-side coverage synthesized alongside them, with AI agents running collection and cross-checking in parallel.
Reconcile
Bases triangulated against each other and restated to a single global scope, so the numbers being compared are actually the same measurement.
Assume, out loud
Penetration is stated as an explicit forward-looking assumption. It belongs to the analyst, so it gets labelled as a judgment rather than dressed up as data.
Deliver in two layers
A clean front-facing report for the reader, over a fully traceable Excel backend where every figure links to its source.
图表 3前台后台两层的研究流程
收集
从多个来源拉取市场基数，同时综合卖方覆盖，由 AI 智能体并行做收集与交叉核对。
统一口径
基数之间互相验证，并统一折算到同一个全球口径，确保比较的是同一种测量。
把假设摆到明面
渗透率写成显式的前瞻假设。它属于分析师的判断，就标成判断，不打扮成数据。
两层交付
给读者一份干净的前台报告，后面压一份完全可溯源的 Excel 后台，每个数字都链到来源。
What is already scheduled or already in motion.
已经排定或已经在跑的事
Exhibit 4Near-term catalysts
| Timing | Event | What it changes | Note |
|---|---|---|---|
| Aug 2026 · done | Greenwoods internship concludes | The market-sizing report, the dashboard and the framework are delivered and reviewable now, not described in the future tense | |
| In development | Course Planner reaches students | First product he owns end to end goes in front of real users. Numbers get published when they exist and not before | |
| Ongoing | Quant system stays in paper trading | Risk discipline gets tested against drawdown before any capital is involved, which is the order he prefers | |
| Jun 2027 | Graduation, B.A. Data Science | Full-time availability begins. No sponsorship required at any point |
图表 4近期催化剂
| 时间 | 事件 | 带来什么变化 | 脚注 |
|---|---|---|---|
| 2026-08 · 已完成 | 景林实习结束 | 测算报告、板块看板与决策框架均已交付，现在就能查验，不是将来时 | |
| 开发中 | 选课助手面向学生 | 他从头负责到尾的第一款产品接触真实用户。有数据了再公布数据，没有就不写 | |
| 持续中 | 量化系统维持 paper trading | 先用回撤检验风控纪律，再谈动用真金白银，这个顺序他不打算颠倒 | |
| 2027-06 | 本科毕业，数据科学学士 | 全职可用性开始 |
The objections, stated first.
先把反对意见摆出来
A risk section that only lists risks nobody would raise is worth nothing. These are the four an experienced hiring manager would actually raise, written the way he would write them about a stock.
只列没人会提的风险，这一节就白写了。下面四条是有经验的招聘负责人真会提的，写法照他写个股的路子。
No full-time buy-side seat yet.
The record is one summer on a desk, not a career. Everything in Section 01 is intern-scope work, produced under a mentor.
MitigantThat summer produced delivered artifacts rather than exposure: a market-sizing report with a traceable backend, a 33-ticker dashboard on a pipeline he built, and a decision framework a senior buy-side investor sat down and reviewed.
Availability is a year out.
Full-time capacity starts Jun 2027. Anything earlier is internship or part-time scope, which does not suit every opening.
MitigantThe date is fixed and on schedule, so it can be planned around rather than guessed at. Work authorization is already in place and needs no sponsorship at any stage.
Heavy reliance on AI tooling.
A large part of his throughput comes from multi-agent workflows. Take the tools away and the output rate drops. That is a real dependency and worth naming.
MitigantThe split is deliberate: agents run collection, cross-checking and scraping; he owns every judgment, every assumption and every number that survives into the report. The systems he builds are constrained to stay explainable to a human reader.
Founder instinct is a retention question.
He started a company at fifteen and is building another product now. A hiring manager is right to ask whether a seat holds someone like that.
MitigantThe builds have always run alongside the day job, not instead of it: Ignite through high school, Course Planner alongside the Greenwoods desk. He runs them as product owner, which means written specs, acceptance criteria and shipped scope.
尚无全职买方经历
记录只有一个夏天，不是一段职业生涯。第 01 节里的产出都是实习生量级，在 mentor 指导下完成。
缓释那个夏天交出来的是成品，不是见习：一份带可溯源后台的测算报告、一套跑在自建管线上的 33 只标的看板、一个被资深买方投资人坐下来复核过的决策框架。
可用时间在一年之后
全职从 2027 年 6 月起算，在那之前只能是实习或兼职量级，未必匹配所有岗位。
缓释毕业时间固定，且按进度推进，可以提前排期，不用猜。
对 AI 工具依赖较重
他的产能有很大一部分来自多智能体流程。工具拿掉，产出速度会掉下来。这是实打实的依赖，得说清楚。
缓释分工是刻意设计的：智能体做收集、交叉核对和抓取；判断、假设，以及最后写进报告的每个数字，由他本人负责。他做的系统也被约束成必须能向人解释清楚。
创业惯性带来的留存问题
15 岁创过公司，现在手上还有产品在做。招聘负责人有理由问一句：一个位子留不留得住这样的人。
缓释这些项目一直是和本职并行，不是替代：Ignite 在高中期间做，选课助手在景林实习期间做。他以产品负责人的方式推进，写规格、定验收、交付范围。
Education
Relevant coursework: Data Science Modeling I & II (STAT 240/340), Data Science Programming (COMP SCI 220), Elementary Matrix & Linear Algebra, Principles of Micro- and Macroeconomics
Professional experience
Greenwoods Asset Management, Shanghai
- Produced a global AI-video industry market-sizing report under a buy-side mentor: multi-source triangulation of market bases with unified global scope, penetration treated as explicit forward-looking assumptions, delivered as a clean front-facing report plus a fully traceable Excel backend where every figure links to its source
- Covered AI-application equities across US, HK and A-share markets: synthesized sell-side research (Goldman Sachs, CITIC-tier coverage) and built a 33-ticker sector dashboard with a self-built market-data pipeline
- Co-developed a 7-step investment decision framework (market selection, thesis with falsification signals, staged position sizing), reviewed by a senior buy-side investor
- Ran an AI-assisted research workflow (multi-agent information collection, cross-checking, data scraping) while owning every judgment and assumption personally
xMEMS, Santa Clara, CA
- Cleaned and transformed finance datasets in Python and loaded them into the accounting system
- Maintained and reconciled accounting records in Excel
- Contributed to a Tableau dashboard visualizing operations and sales performance
- Assisted SEO keyword analysis (SEMrush) informing website improvements
Ignite 3D Hologram Co.
- Designed a 3D hologram device; secured a national utility-model patent and a design patent
- Pitched the business plan to two venture investors and iterated on their feedback
Projects
- Building a web product that turns a student's DARS degree audit into three or fewer ranked, conflict-free semester schedules: upload the audit, pick a preference profile, get complete schedules with explainable trade-offs for every course and professor choice
- Unified three heterogeneous data sources (Madgrades grade-distribution API, UW Course Search & Enroll, RateMyProfessor) into one data model; Python/FastAPI backend with a constraint-based recommendation engine
- Runs the build as product owner with AI agents as the engineering team: owns spec, roadmap and acceptance, delegates implementation to multi-agent workflows
- Designed an end-to-end daily pipeline in Python: interpretable stock-selection model on price, volatility, drawdown and volume features; logistic regression for next-day move probabilities
- Cross-sectional ranking selects a concentrated 3 to 5 stock portfolio from a roughly 60-ticker high-liquidity universe, under realistic constraints (integer shares, cash buffer, sizing rules)
- Built data validation and freshness checks to prevent look-ahead bias; evaluated via paper trading with drawdown-focused risk control
- Used LLMs to translate investment constraints into rule-based logic and stress-test features, keeping the whole system explainable
Skills
Python (pandas, NumPy, scikit-learn), SQL, Tableau, Advanced Excel. LLM-powered workflows: multi-agent orchestration, prompt engineering, AI-assisted research pipelines.
Interests
Vocal performance (6 years of training, top-4 in university competition) · data-driven fitness training (5 years) · badminton · poker
Status
Authorized to work in the U.S. without sponsorship.
教育背景
相关课程：数据科学建模 I & II（STAT 240/340）、数据科学编程（COMP SCI 220）、矩阵与线性代数、微观与宏观经济学原理
工作经历
景林资产，上海
- 在买方 mentor 指导下完成全球 AI 视频行业市场规模测算报告：多来源交叉验证市场基数并统一全球口径，渗透率作为显式的前瞻假设处理；交付形式是一份干净的前台报告，加一份完全可溯源的 Excel 后台，其中每个数字都链回来源
- 覆盖美股、港股、A 股的 AI 应用类公司：综合高盛、中信级别的卖方研究，并在自建行情数据管线上搭建 33 只标的的板块看板
- 共同搭建 7 步投资决策框架（选市场、带可证伪信号的投资命题、分步建仓），经资深买方投资人复核
- 运行 AI 辅助的研究流程（多智能体信息收集、交叉核对、数据抓取），所有判断与假设由本人负责
xMEMS，美国加州圣克拉拉
- 用 Python 清洗与转换财务数据集，并导入账务系统
- 在 Excel 中维护账务记录并完成对账
- 参与搭建反映运营与销售表现的 Tableau 看板
- 协助 SEO 关键词分析（SEMrush），为网站改版提供依据
Ignite 3D 全息
- 设计 3D 全息设备，获国家实用新型专利与外观设计专利各一项
- 向两位风险投资人路演商业计划，并按其反馈迭代
项目经历
- 在做的产品把学生的 DARS 学位审计转成不超过 3 套排序过的无冲突学期课表：上传审计、选一个偏好档案，拿到完整课表，每门课和每位教授的取舍都给得出理由
- 把三个异构数据源（Madgrades 成绩分布 API、UW 选课系统 Course Search & Enroll、RateMyProfessor）统一成一套数据模型；Python/FastAPI 后端配基于约束的推荐引擎
- 以产品负责人的身份推进，AI agent 团队充当工程团队：规格、路线图与验收由他把关，实现委托给多智能体流程
- 用 Python 搭建端到端日频流水线：基于价格、波动、回撤与成交量特征的可解释选股模型，以 logistic regression 估计次日涨跌概率
- 横截面排序从约 60 只高流动性标的池中选出 3 到 5 只集中持仓，并施加现实约束（整数股、现金缓冲、仓位规则）
- 建立数据校验与新鲜度检查，防止用到未来数据；以 paper trading 评估，风控以回撤为核心
- 用 LLM 把投资约束翻译成规则逻辑并压力测试特征，保持整套系统可解释
技能
Python（pandas、NumPy、scikit-learn）、SQL、Tableau、高级 Excel。LLM 工作流：多智能体编排、prompt engineering、AI 辅助研究管线。
兴趣
声乐（6 年训练，校级比赛前四）· 数据驱动的健身训练（5 年）· 羽毛球 · 德州扑克
状态
Authorized to work in the U.S. without sponsorship.
Every figure in this report, and where it comes from.
报告里的每个数字，以及它的出处
This appendix is the backend. Each note carries the claim, the role it came from and the detail behind it. Open them from the report or clear them here.
这一节就是后台。每条注给出论断、它来自哪段经历，以及背后的细节。可以从正文点开，也可以在这里逐条清掉。
-
1
Global AI-video market-sizing report, delivered clean at the front and traceable at the back
全球 AI 视频行业市场规模测算报告：前台干净，后台可溯源
Produced under a buy-side mentor. Delivered as a clean front-facing report plus a fully traceable Excel backend in which every figure links to its source.
在买方 mentor 指导下完成。交付形式是一份干净的前台报告，加一份完全可溯源的 Excel 后台，其中每个数字都链回它的来源。
-
2
Sizing method: triangulation, one global scope, penetration stated as an assumption
测算方法：多源交叉验证、统一全球口径、渗透率作为显式假设
Multi-source triangulation of market bases restated to a unified global scope, with penetration treated as an explicit forward-looking assumption rather than folded into the base.
多来源交叉验证市场基数并统一折算到全球口径；渗透率作为显式的前瞻假设处理，不揉进基数里。
-
3
33-ticker sector dashboard on a self-built market-data pipeline
33 只标的的板块看板，跑在自建行情数据管线上
Built to track the AI-application coverage universe. The dashboard sits on a market-data pipeline he built rather than a vendor terminal feed.
用于跟踪 AI 应用类的覆盖池。看板底下是他自己搭的行情数据管线，不是终端现成的数据源。
-
4
Coverage of AI-application equities across US, HK and A-share markets
覆盖美股、港股、A 股的 AI 应用类公司
Synthesized sell-side research at the Goldman Sachs and CITIC tier across three markets into one coverage view.
综合高盛、中信级别的卖方研究，把三个市场并进同一套覆盖视角。
-
5
7-step investment decision framework, reviewed by a senior buy-side investor
7 步投资决策框架，经资深买方投资人复核
Co-developed on the desk. Covers market selection, a thesis carrying explicit falsification signals, and staged position sizing.
在研究台上共同搭建，覆盖选市场、带显式可证伪信号的投资命题、分步建仓。
-
6
AI-assisted research workflow with personal ownership of every judgment
AI 辅助的研究流程，判断与假设由本人负责
Multi-agent information collection, cross-checking and data scraping ran the legwork. Every judgment and assumption in the output was owned personally.
多智能体负责信息收集、交叉核对与数据抓取这些跑腿的活。产出里的每一个判断和假设由本人认领。
-
7
B.A. in Data Science, University of Wisconsin-Madison, expected Jun 2027
威斯康星大学麦迪逊分校数据科学学士，预计 2027 年 6 月毕业
Relevant coursework: Data Science Modeling I & II (STAT 240/340), Data Science Programming (COMP SCI 220), Elementary Matrix & Linear Algebra, Principles of Micro- and Macroeconomics.
相关课程：数据科学建模 I & II（STAT 240/340）、数据科学编程（COMP SCI 220）、矩阵与线性代数、微观与宏观经济学原理。
-
8
Degree audit in, three or fewer conflict-free schedules out
输入学位审计，输出不超过 3 套无冲突课表
A student uploads a DARS degree audit, picks a preference profile and gets complete ranked schedules with explainable trade-offs for every course and professor choice.
学生上传 DARS 学位审计、选一个偏好档案，拿到排序过的完整课表，每门课和每位教授的取舍都给得出理由。
-
9
Three heterogeneous data sources unified into one model
三个异构数据源统一成一套模型
Madgrades grade-distribution API, UW Course Search & Enroll and RateMyProfessor reconciled into a single data model behind a Python/FastAPI backend with a constraint-based recommendation engine.
Madgrades 成绩分布 API、UW 选课系统 Course Search & Enroll 与 RateMyProfessor 统一成一套数据模型，后端是 Python/FastAPI 加基于约束的推荐引擎。
-
10
Run as product owner with AI agents as the engineering team
以产品负责人身份推进，AI agent 团队充当工程团队
He owns spec, roadmap and acceptance. Implementation is delegated to multi-agent workflows rather than written by hand.
规格、路线图与验收都在他手里；实现委托给多智能体流程，而不是自己一行行写。
-
11
Interpretable selection model, 3 to 5 names from a ~60-ticker universe
可解释选股模型，从约 60 只标的池中选 3 到 5 只
Features on price, volatility, drawdown and volume; logistic regression for next-day move probabilities; cross-sectional ranking into a concentrated portfolio under integer-share, cash-buffer and sizing constraints.
特征取价格、波动、回撤与成交量；用 logistic regression 估计次日涨跌概率；横截面排序形成集中持仓，受整数股、现金缓冲与仓位规则约束。
-
12
Look-ahead checks, paper trading, drawdown-focused risk control
防未来数据、paper trading、以回撤为核心的风控
Data validation and freshness checks prevent look-ahead bias. Evaluation runs in paper trading. LLMs translate investment constraints into rule-based logic so the system stays explainable.
数据校验与新鲜度检查防止用到未来数据。评估在 paper trading 中进行。LLM 把投资约束翻译成规则逻辑，系统因此保持可解释。
-
13
Finance data engineering and reporting at a hardware company
硬件公司里的财务数据工程与报表
Cleaned and transformed finance datasets in Python and loaded them into the accounting system, maintained and reconciled records in Excel, contributed to a Tableau dashboard on operations and sales, and assisted SEO keyword analysis in SEMrush.
用 Python 清洗与转换财务数据并导入账务系统，在 Excel 中维护与对账，参与运营与销售的 Tableau 看板，并协助 SEMrush 的 SEO 关键词分析。
-
14
A company at fifteen, two patents, two venture pitches
15 岁的一家公司、两项专利、两次风投路演
Designed a 3D hologram device and secured a national utility-model patent plus a design patent. Pitched the business plan to two venture investors and iterated on their feedback.
设计 3D 全息设备，拿到一项国家实用新型专利和一项外观设计专利。向两位风险投资人路演商业计划，并按其反馈迭代。
-
15
Toolchain
工具栈
Python (pandas, NumPy, scikit-learn), SQL, Tableau, advanced Excel. LLM-powered workflows: multi-agent orchestration, prompt engineering, AI-assisted research pipelines.
Python（pandas、NumPy、scikit-learn）、SQL、Tableau、高级 Excel。LLM 工作流：多智能体编排、prompt engineering、AI 辅助研究管线。
-
16
Authorized to work in the U.S. without sponsorship
美国工作授权
No sponsorship is required at any stage of hiring. The analyst on this report and the subject of it are the same person, which is disclosed in Section 09.
Authorized to work in the U.S. without sponsorship. 本报告的分析师与覆盖标的为同一人，已在第 09 节披露。
I, Chaoran Jiang, certify that the views expressed in this report accurately reflect my personal record, and that no part of my compensation was, is, or will be tied to overstating it.
Chaoran Jiang · 21 August 2026 · Madison, WI and Hangzhou, China
- ConflictThe analyst and the covered subject are the same person. That is the entire conflict of interest, and it is disclosed here rather than buried on the last page. The mitigant is the one the desk taught him: every figure in this report carries a source note, and the notes are the record.
- ValuationThis report contains no price target, no earnings estimate and no valuation. The only forward-looking item in it is availability, and that date is Jun 2027.
- RatingsBUY: worth an interview. HOLD: keep the file open. SELL: you are hiring for something this record does not cover, which is a fine outcome for both sides.
- ScopePrepared for hiring managers in equity research, investment analytics and adjacent seats. Not investment advice, and not a substitute for asking him hard questions in person.
本人蒋超然声明：本报告所述观点准确反映了本人的真实记录；本人的任何报酬，过去、现在或将来，均不与夸大这份记录挂钩。
蒋超然 · 2026 年 8 月 21 日 · 麦迪逊 / 杭州
- 利益冲突本报告的分析师与覆盖标的是同一个人。这就是全部的利益冲突，写在这里，而不是塞进最后一页。缓释办法用的是研究台教他的那一套：报告里每个数字都挂着溯源脚注，脚注本身就是记录。
- 估值本报告不含目标价、不含盈利预测、不含任何估值。唯一的前瞻项是可用时间，日期为 2027 年 6 月。
- 评级定义BUY：值得面一次。HOLD：档案先留着。SELL：你要招的岗位这份记录覆盖不到，这对双方都是好结果。
- 适用范围面向股票研究、投资分析及相关岗位的招聘负责人。不构成投资建议，也替代不了当面问他几个难题。
Coverage inquiries
联系分析师
Rating is BUY and the target is a single seat. If the fit looks right, the fastest next step is a conversation.
评级 BUY，目标是一个席位。方向合适的话，最快的下一步是聊一次。