I work across quantitative modeling, software engineering, and rule-heavy domains such as tax, legal processes, and finance. I studied Economics and Applied Statistics at the University of Michigan, and later worked in transfer pricing at PwC Shanghai. Much of my work has focused on turning manual, document-heavy workflows into tools that are easier to check, reproduce, and maintain.
This site is a small portfolio of things I have built or written: report QC tools, German tax filing automation, bank-statement parsers for intercompany lending, PE/VC transaction documents, and a research paper that formulates part of the Pillar Two tax regime as an optimization problem. Some projects are formal work products; others are experiments, notes, or public-facing demos.
I am interested in roles where code, quantitative reasoning, and domain structure matter — including asset management, fintech infrastructure, legal tech, and quantitative finance. I also have working knowledge of machine learning and time-series modeling; I have not built as many public projects in that area, but the methods are within my comfort zone. I am also open to work involving LLM application debugging or evaluation, with some time to ramp up on the specific stack and task context.
密歇根大学经济学和应用统计出身，后来在普华永道上海做转让定价。比较关注量化建模、软件工程，以及税务、法律、金融这类规则密集型领域之间的交叉。
这个页面先当个人作品集用：放我做过或写过的东西，包括报告质检工具、德国 CbCR 自动申报、银行对账单解析、PE/VC 交易文件，以及把 Pillar 2 税制写成优化问题的研究。以后一些 demo、笔记、视频主页也可以往这里放。
现在主要看资管、金融基础设施、法律科技、量化相关方向。机器学习和时间序列也可以做，只是目前公开项目没有放太多；方法本身不算陌生。大模型应用调试、评估相关的工作也可以做，但需要一定时间熟悉具体业务场景和工程栈。
End-to-end pipeline for Country-by-Country Reporting to the German BZSt. Reconciles multi-entity, multi-currency financial data, generates XML Schema 2.0 compliant files, handles the ELMA5 → DIP portal migration, DocRefId formatting, Steuernummer validation. In production for live client filings. 给德国联邦税务局（BZSt）做的国别报告全流程自动化。多实体、多币种的财务数据勾稽，生成 XML Schema 2.0 的申报文件，ELMA5 到 DIP 的系统切换、DocRefId、税号校验这些坑全踩过。已经在客户实际申报里跑了。
Desktop app that uses LLMs to review transfer pricing reports — catches factual errors, number mismatches, and language issues. Also batch-rewrites outdated paragraphs in annual updates (200+ per run), preserving Word formatting. 接了大模型的桌面工具，查转让定价报告里的事实错误、数据对不上和语言问题。也能批量改写年度更新的过期段落，一次跑 200 多段，Word 格式不会乱。
Parses five bank-statement formats across three cash pools, computes related-party lending, occupied days, annualized interest. One-click output. 五种银行对账单格式自动解析，三个资金池的关联方借贷、占用天数、年化利息一把算完，点一下出表。
Drafted full sets of Series A transaction agreements — capital increase, equity transfer, shareholders' agreement — with cross-referencing provisions, anti-dilution clauses, and OOXML tracked-changes markup for collaborative review. Comfortable working across the boundary between legal structure and financial modeling. 写过 Series A 全套文件——增资协议、股转协议、股东协议，反稀释、交叉引用都有，OOXML 修订标记走的协作审阅。