我把真实世界里模糊、琐碎又反复发生的问题，做成真正有人使用的产品。
I turn ambiguous real-world problems into working digital products.

我做过企业内部工具、面向普通用户的产品，也会做一些解决自己问题的小工具和有趣的数字作品。我最擅长的是走进一个真实场景，弄清楚大家到底卡在哪里，再把混乱的流程梳理清楚、做成产品，并根据使用反馈继续改进。

My work spans enterprise AI workflows, consumer products, and small creative experiments. I enjoy the work between a messy operating process and a tool people can actually use: understanding the domain, structuring the workflow, building the product, and iterating with users.

这里收录的是我为真实企业和社会机构落地的数字化项目。出于保密要求，公开仓库均为脱敏展示版，不包含生产源码、账号凭证、真实机构名称或敏感业务数据。
Real systems built around real organizational workflows. Public repositories are sanitized case studies: production code, credentials, organization identities, and sensitive data remain private.

这五个项目都由我主导完成从 0 到 1 的落地，包括理解业务、梳理需求、设计方案、开发实现、测试交付和后续迭代。项目负责人和管理者提供业务背景与管理视角，一线使用者持续反馈真实问题；我负责把这些信息转化为具体产品并推进落地。

I led all five projects from zero to one, covering workflow discovery, requirements, product design, implementation, delivery, and iteration. Program leads, managers, and frontline users contributed domain context and feedback; I turned that input into working products.

项目服务于残障青年职业发展计划。教师原本每学期都要手工完成 100 多项指标的评估、算分、制图和报告撰写，零散材料也很难呈现同一名学生几年来的成长变化。我把专业评估方法做成线上评分、自动算分与报告生成系统，并为每位学生建立连续档案，支持对比在校期间的多轮评估。第一版完成后，我通过教师试用和培训持续调整评分顺序、指标说明、报告呈现与档案查找方式，最终完成发布与交付。

Built for a vocational development program serving young people with disabilities. Teachers previously evaluated 100+ indicators, calculated scores, created charts, and wrote reports by hand, while fragmented records made long-term progress difficult to track. I turned the professional framework into an online assessment, automated reporting, and longitudinal student record system, then refined the workflow, explanations, reports, and record navigation through teacher testing and training.

公司原本把产品、配件和价格分散在多个 Excel 表格里。销售需要反复查找、复制和核价，再依赖售前经验确认配置，报价格式和历史版本也难以统一。我从销售、售前和管理者的实际流程出发，建立统一产品数据库，将客户信息、产品选配、价格、预览、Excel 导出和历史报价串成一套系统；后续又根据实际报价反馈补充搜索、配件关系、账号权限和后台维护能力，让报价从个人经验驱动变成可维护、可追踪的标准流程。

Product, accessory, and pricing data had been scattered across spreadsheets. Salespeople repeatedly searched, copied, and verified information, relied on presales expertise for configurations, and lacked consistent quotation formats or version history. I built a centralized product database and connected customer details, configuration, pricing, preview, Excel export, and quotation history into one workflow, then added search, accessory rules, permissions, and administration based on real usage.

这个项目最初只是为经理完成一次月度销售统计，但我发现客户和项目动态长期散落在销售日报、周会纪要和个人记忆里，每个月重新整理既重复，也看不到一个商机从出现到推进、停滞或结束的完整过程。我把一次性统计扩展成持续更新的客户与商机管理系统，统一记录负责人、阶段、进展、下一步行动和停滞原因，并用看板呈现商机总量、实际推进项目与推进率。经理提供复盘视角，我据此持续调整状态定义、指标口径和看板层级。

This began as a one-off monthly sales summary. I recognized that customer and opportunity updates were repeatedly reconstructed from daily reports, meeting notes, and memory, with no continuous view from first contact through progress, stagnation, or closure. I productized the task into an ongoing opportunity system with ownership, stage, progress, next actions, stagnation reasons, and management dashboards, iterating its definitions and metrics with the manager.

团队过去需要每天在招标网站用多组关键词人工搜索，再逐条判断相关性、招标阶段和跟进价值。单一关键词容易漏项，多组关键词又会产生大量噪音，而且“教育局发布但业务关联较弱的项目也值得保留”这类判断无法靠简单关键词表达。我搭建了批量抓取、AI 阅读、业务规则筛选、优先级与阶段分类、特殊主体保留和每日简报的完整流程。早期通过经理人工标注与 AI 结果逐条对照，不断修正规则和 Prompt，才把经验沉淀成稳定的日常工作流。

The team previously searched tender sites with multiple keywords every day and manually judged relevance, stage, and follow-up value. A single query missed opportunities, while broader retrieval created heavy noise and could not express exceptions such as retaining education-bureau notices for relationship building. I built the full workflow from batch retrieval and AI reading to business-rule filtering, priority and stage classification, exception handling, and daily briefs, calibrating it through repeated comparison with manager-labeled examples.

最初的会议纪要 Skill 能统一格式，却会写错逐字稿里的新客户、学校和项目名称，甚至把工作对应到错误的销售人员。为了让输出真正可用，我找到 CRM 销售日报这一内部信息源，主动协调每周导出，并让系统用日报核对逐字稿中的名称与人员关系；无法确认的内容会明确提示，而不是自行补全。经过每周真实使用和持续修正规则后，原来需要多方逐项核对的纪要，缩短为生成后快速确认即可提交。

The first version standardized meeting-note structure but still misidentified unfamiliar customers, schools, projects, and their owners. I introduced weekly CRM sales reports as an internal source of truth, coordinated their export, and designed cross-checks for entity names and ownership, flagging uncertainty instead of inventing details. Repeated weekly use and rule refinement turned a multi-person verification task into a fast final review.

与伙伴共同打造的产品；生产源码保持私有，通过公开展示仓库介绍产品与过程。
Collaborative products with private production source code and public product showcases.

WriteBuff 是与伙伴共同开发的 IELTS/TOEFL AI 桌面写作伙伴。它不要求学习者离开 Word、Google Docs 等熟悉的写作环境，而是通过桌面伙伴和快捷键把获取题目、计时写作、提交作文、查看评分与逐句修改、保存表达和定期复习连成完整练习闭环。我主要负责 UI、用户体验和前端页面设计，需要同时处理桌宠的轻量陪伴、复杂报告的信息层级、提醒不过度打扰，以及剪贴板读取过程中的隐私与信任提示。

WriteBuff is a collaborative AI desktop writing companion for IELTS and TOEFL learners. It preserves familiar writing environments while connecting prompts, timed practice, submission, feedback, sentence-level revision, expression saving, and spaced review into one loop. I lead UI, UX, and frontend interface design, balancing a lightweight desktop companion with information-dense reports, non-disruptive reminders, and clear privacy cues.

MadeSure 是与伙伴共同开发、已经公开上线的中国供应商尽调产品。海外采购者输入供应商营业执照上的统一社会信用代码后，系统会确认注册主体，交叉整理企业登记、贸易、平台、法律与其他公开信息，并生成带来源链接、核实状态和信息边界的报告。我主要负责前端设计，把复杂的研究流程转化为用户能理解的操作路径和报告结构，重点处理证据层级、风险提示、事实与推断的区分，以及官网从产品解释到发起核查的转化路径。

MadeSure is a live, collaborative due-diligence product for overseas buyers evaluating Chinese suppliers. Starting from a supplier's Unified Social Credit Code, it verifies the registered entity, cross-checks public business, trade, marketplace, legal, and other evidence, and produces a source-linked report that distinguishes verification status and limitations. I lead frontend design, translating the research workflow into understandable interactions and reports with clear evidence hierarchy, risk communication, and conversion paths.

这一部分的项目都由我独立完成，从最初的想法、产品与交互设计，到开发实现、测试和公开发布。

I independently took each project in this section from idea and product design through implementation, testing, and release.

我在用 ChatGPT 研究和写作时，发现有价值的内容留在对话里难以重新组织，复制到笔记软件又会失去上下文，于是独立做了这个 Chrome 侧栏插件。它支持把对话内容一键摘录到白板或文档，继续拖拽、连线、标注和改写，并能定位回原文或把整理后的内容再次交给 ChatGPT；数据优先保存在本地浏览器中。

I independently built this Chrome side-panel extension to bridge the gap between useful ChatGPT conversations and structured thinking. Users can capture passages into boards or documents, organize and connect them, return to the source, and send refined material back to ChatGPT, with local-first storage.

我希望收藏动作足够轻，同时又不让内容长期堆在收件箱里，因此独立设计了一套分阶段工作流：手机端通过快捷指令快速保存文字、链接、图片和个人备注到 Obsidian；Codex 每天只读取新增材料，结合图片语境完成去重、归类，并生成带行动项、研究问题和来源索引的知识简报。整个流程由我独立完成，包括快捷指令、Skill、处理脚本和文档。

I independently designed a two-stage capture workflow: an iPhone Shortcut quickly saves text, links, images, and notes to Obsidian, while Codex incrementally processes only new material into deduplicated, categorized daily briefs with actions, research questions, and source indexes. I built the Shortcut, Skill, scripts, and documentation.

这个项目解决的不是“把照片放进模板”，而是怎样真正讲出一个具体女性的故事。我把它做成一个会先访谈、再策展的 Codex Skill：逐步了解人物、关系与经历，共同确定叙事结构，再完成文字、统一画风、轻动画、音乐和页面，最终生成可离线保存与转发的单文件 HTML。我独立设计了访谈与策展方法、交互模板、构建脚本、隐私规则和示例案例。

This project is about telling one woman's specific story rather than placing photos into a generic template. I independently built a Codex Skill that interviews the user, develops a narrative structure, and then assembles writing, a consistent visual style, light animation, music, and interaction into a private, shareable offline HTML gallery.

为了让没有读过《奥德赛》的观众在观影前快速理解人物、事件和主题，我把传统长文导读改成第一人称互动旅程。用户以奥德修斯的身份沿地图推进，在独眼巨人、喀耳刻、冥界、塞壬等节点作出选择，并在最后回看自己的归乡记录。我独立完成叙事结构、关卡与选择、视觉方向、交互实现和演示构建。

I independently turned a conventional pre-film guide into a first-person interactive voyage. Users travel as Odysseus, make choices across key episodes, and build a personal homecoming record, learning the characters, events, and themes through spatial narrative rather than a long summary.

在线体验 / Live Demo · 项目仓库 / Repository

传统纸质族谱和长篇资料不便检索，也很难让年轻人理解自己与家族历史的关系。我独立设计了一个移动端产品概念，把族谱检索、支派筛选、关系图、家族文化、新闻、宗亲活动和个人入口放在同一套信息架构中，探索族谱从静态档案变成代际连接与家族社区入口的可能性；当前仓库展示的是脱敏界面与交互原型。

I independently designed this mobile product concept to make family records searchable and relevant to younger generations. It combines genealogy search, branch filters, relationship maps, cultural content, news, community activities, and personal entry points, exploring how a static archive could become a tool for intergenerational connection.

业务理解 Business context → 流程设计 Workflow design → 快速构建 Rapid build → 用户反馈 User feedback → 迭代交付 Iteration & delivery

我会根据交付问题选择技术，而不是把技术栈本身当成产品。当前项目涉及 TypeScript、React、Python、Tauri、SQLite、Supabase、LLM API 与 Agentic Coding 工作流。
I choose the stack around the delivery problem rather than treating the stack as the product. Current projects use TypeScript, React, Python, Tauri, SQLite, Supabase, LLM APIs, and agentic coding workflows.