张紫琼 · 产品经理作品集 | Zhang Ziqiong's Portfolio

门店智能开单与账单管理系统 Smart Order & Billing Management System

AI-Assisted Retail Order Management System · Hengtong Fabrics

独立项目 (Personal Project)Independent Project 2026.03 – 至今Mar 2026 – Present 产品设计 + AI 辅助全栈验证Product Design + AI-Assisted Full-Stack Validation

GitHub 源码 ↗GitHub Source ↗

痛点分析 Problem Statement Problem Statement

经实地走访，所在的大型家具与建材市场中（涵盖纺织布艺、皮革、五金等多种业态），近 2/3 的商户受限于相似的复杂客情，仍在使用纯人工记账与手写开单。

Through field research at large furniture and building material markets (covering textiles, leather, hardware, etc.), we found that nearly 2/3 of merchants are still using purely manual bookkeeping and handwritten receipts due to similar complex customer relationships.

①

老客户信息重复填写，历史单价无法复用，每次都要翻旧货本才能找到对应价格。

Repeated Entry: Historical unit prices cannot be reused. Every time, merchants must flip through old ledgers to find the corresponding price.

②

"千人千价"完全靠人脑记忆，高峰期多线并行时极易出现价格错误，造成额外赔偿损失。

"Thousand People, Thousand Prices": Relies entirely on human memory. During peak hours, multi-tasking easily leads to pricing errors, causing financial compensation losses.

③

欠账记录混入手写单，繁忙时段容易漏记忘记，管理层无法实时汇总查看应收款。

Missing Records: Debt records are mixed with handwritten receipts, making them easy to forget during busy periods. Management cannot view accounts receivable in real-time.

15–40 每日处理订单数
（峰值 40+ 单/天） Daily processed orders
(Peak 40+/day)

2–5min 原平均单张开单耗时
（含查价、手写、拍照） Avg. time per order
(incl. checking price & writing)

1–2次/月/mo 客户信息或金额错误
频率（真金白银的代价） Frequency of price errors
(Real financial loss)

2/3+ 家具材料零售中小型
商户的共性痛点 Common pain point for
SME retail merchants

业务壁垒与竞品分析 Competitive Analysis Competitive Analysis

🚧 极度非标的 SKU 与复杂的客情定价 🚧 Highly non-standard SKUs & complex pricing

纺织布料等家具材料的颜色、花纹、薄厚高度依赖不同批次的工厂出货，型号繁多（经调研，当前门店拥有高达 13,500+ 种不同型号）。且由于包含大量长期与散客，加上随行就市的同行竞争，形成了绝对的“千人千价”。

Textile fabrics and furniture materials depend heavily on batch variations. Models are numerous (the current store has over 13,500+ different SKUs). Combined with long-term clients, retail customers, and dynamic market competition, an absolute "Thousand People, Thousand Prices" scenario exists.

⚔️ 竞品分析与破局差异化 ⚔️ Competitive Analysis & Differentiation

❌ 通用网店 / 餐饮点单系统（如微商城、点单SaaS）：强制要求商家预先录入标准化的商品图文与固定单价，完全无法适应纺织批发 13,500+ 非标 SKU 的变动，以及灵活的改价场景。
❌ 传统进销存 ERP 系统：前台移动端录入极度繁琐，无法顺滑衔接“客户通过微信发送大段文字报单”的行业习惯。
✅ 本系统的破局点：抛弃沉重死板的商品库包袱，采用“轻量级输入解析 + 客户关系单价映射”的模式。核心只维护“客户-价格”的逻辑链，打造出最符合传统建材批发业态的“极简开单流”。

❌ General E-commerce / POS Systems: Require merchants to pre-enter standardized product images and fixed unit prices. Completely fails to adapt to 13,500+ non-standard SKUs and flexible pricing scenarios.
❌ Traditional ERP Systems: Mobile entry is extremely cumbersome. Cannot smoothly connect with the industry habit of "customers sending long text orders via WeChat".
✅ Our Breakthrough: Abandoning the heavy product library baggage. Adopting a "lightweight input parsing + customer relationship price mapping" model. Only maintaining the "customer-price" logical chain, creating the most suitable "minimalist order flow" for the traditional wholesale industry.

目标用户 User Persona User Persona

👩

周姐 · 资深店员Sister Zhou · Senior Clerk

50+ 岁 · 主力接单、开单负责人Age 50+ · Main order handler

日常场景Daily Scenarios

全天通过手机微信接收客户报单，手写销货单，人工计算金额，拍照发客户，整理欠账。

Receives orders via WeChat all day, handwrites receipts, calculates amounts manually, takes photos for customers, organizes debts.

核心痛点Core Pain Points

高峰期多线并行，价格靠记忆，出错要赔偿。每次都需翻旧单子查老客价格，耗时费力。

Multi-tasking during peak hours, relies on memory for pricing, errors lead to compensation. Flipping through old receipts is time-consuming.

目标Goals

开单快、不出错、不用翻旧本子，历史价格自动带出。

Fast ordering, error-free, automatic retrieval of historical prices.

👴

张老板 · 店主Boss Zhang · Store Owner

50+ 岁 · 监督账单 + 偶尔开单Age 50+ · Billing supervisor

日常场景Daily Scenarios

不定期协助开单，重点关注欠款情况，需要核实客户账目，偶尔查阅历史单据处理纠纷。

Occasionally helps with orders, focuses on debt tracking, verifies accounts, checks historical documents for disputes.

核心痛点Core Pain Points

欠账汇总只能人工盘点，看不到实时数据；多人共享数据靠口头传达，容易遗漏。

Debt summaries rely on manual counting without real-time data. Sharing data relies on verbal communication, prone to omissions.

目标Goals

随时查看欠账总览，多端成员共享同一份实时数据库。

Real-time overview of debts, real-time shared database across multiple devices.

用户旅程优化 User Journey User Journey Optimization

⚠ Before — 原手写开单流程 (7步) ⚠ Before — Original Handwritten Flow (7 Steps)

微信报单WeChat Order

确认型号数量Confirm SKU & Qty

易遗漏Easy to miss

查找历史价格Check Past Price

翻旧本耗时Time-consuming

手写销货单Write Receipt

重复抄写Repetitive

手工计算金额Manual Calc

容易算错Prone to errors

拍照发客户Photo to Client

效率低下Low efficiency

手记欠账Note Debt

繁忙易漏Easy to forget

✓ After — 数字化流程 (仅需 3 步人工操作) ✓ After — Digital Flow (Only 3 manual steps)

1. 粘贴报单文本1. Paste Text

简单输入客户姓名Input customer name

⚡ 自动解析与价格匹配⚡ Auto Parse & Match

2. 确认并一键生成2. Confirm & Generate

校验无误直接出单One-click generation

☁️ 自动同步云端账单☁️ Auto Cloud Sync

3. 转发发送客户3. Share to Client

导出图片或直接分享Export image or share

用户故事 User Stories User Stories

用户角色 (As a...)User Role (As a...)	需求描述 (I want to...)Requirement (I want to...)	核心价值 (So that...)Value (So that...)
店员Clerk	直接粘贴微信报单文本，系统自动解析客户信息、型号和数量。	Directly paste WeChat order texts, let the system automatically parse customer info, SKUs, and quantities.	彻底免去手抄信息的繁琐，从源头避免抄写错误。	Completely eliminate the hassle of manual copying and prevent transcription errors at the source.
店员Clerk	系统能自动检索老客专属价格，匹配不到时再兜底默认价格。	The system can automatically retrieve exclusive prices for returning customers, falling back to default prices if unmatched.	消除“千人千价”的记忆负担，告别翻旧账本，规避算错账赔偿风险。	Eliminate the memory burden of dynamic pricing, say goodbye to old ledgers, and avoid financial compensation risks.
店员Clerk	将数据一键渲染成发货单图片并支持微信分享。	Render data into a receipt image with one click and support WeChat sharing.	可以直接转发给客户确认，省去举着手机找光线拍照的尴尬。	Can directly forward to clients for confirmation, saving the awkwardness of taking photos with poor lighting.
店主Owner	拥有独立工作台，能实时查看全盘欠账总计与流水明细。	Have an independent dashboard to view real-time total debts and transaction details.	随时掌握资金流向与应收账款，告别期末人工对账的痛苦。	Monitor cash flow and accounts receivable anytime, saying goodbye to the pain of manual reconciliation.
全员All Staff	多台设备同时登录系统，数据基于云端即时同步。	Log into the system from multiple devices simultaneously, with data instantly synced via the cloud.	团队成员不再孤立作战，库存与账目信息差彻底拉平。	Team members no longer work in silos, completely eliminating information gaps in inventory and accounts.

核心功能模块 Key Features Key Features

📝 智能录单与订单编辑Smart Order Entry & Editing

系统支持将门店日常收到的报单文本快速整理为可编辑订单。

在录入后，页面会自动识别客户、商品、数量等信息，并结合客户资料、客户专属价和默认价进行补全，帮助店员从“收到报单”快速进入“可确认开单”的状态。

The system supports quickly organizing daily order texts received by the store into editable orders.

Upon entry, the page automatically recognizes customer info, SKUs, and quantities. It then auto-fills using customer profiles, exclusive prices, and default prices, helping clerks swiftly transition from "receiving orders" to "confirming orders".

文本转结构化订单Text to Structured Order 自动补全客户资料Auto-fill Profiles 客户价 / 默认价优先匹配Smart Price Matching 移动端友好的编辑体验Mobile-friendly Editing

🧾 销货单生成与图片化导出Receipt Generation & Image Export

确认订单后，系统会生成标准化销货单预览，并支持图片导出、复制、下载和分享。

这样可以把原本零散的报单内容，快速转成适合发给客户、店内留档或继续流转的标准单据，提升门店日常开单效率。

After confirming the order, the system generates a standardized receipt preview, supporting image export, copy, download, and sharing.

This quickly converts scattered order content into standard documents suitable for sending to customers, in-store archiving, or further processing, boosting daily efficiency.

标准化销货单模板Standard Receipt Template 图片导出与分享Image Export & Share 适合移动端使用Mobile Optimized 从录单到交付一体化完成End-to-end Delivery

⏳ 历史记录管理与重生成History Management & Regeneration

系统支持保存历史销货单，并提供搜索、分页查看、详情展开和再次编辑重生成能力。

店员可以快速回看过去的开单记录，并基于原有数据继续修改，而不需要每次从头录入，提升重复开单和改单场景下的操作效率。

The system supports saving historical receipts, offering search, pagination, detailed views, and the ability to re-edit and regenerate.

Clerks can quickly review past records and modify them based on original data without starting from scratch, highly improving efficiency for repeated or modified orders.

历史销货单留存Receipt History Retention 搜索与分页查看Search & Pagination 详情展开Detail Expansion 继续编辑并重生成Re-edit & Regenerate

🗂️ 客户资料与价格规则管理Client Profiles & Pricing Rules

系统内置业务数据库管理能力，支持维护客户信息、客户专属价格和默认价格表。

通过把价格规则和客户资料沉淀下来，后续开单时可以自动带出常用信息，减少重复输入，也让门店报价逻辑更统一、更易维护。

Built-in business database capabilities support maintaining customer info, exclusive pricing, and default price lists.

By accumulating pricing rules and profiles, future orders automatically fetch common info, reducing repetitive input and making the store's pricing logic unified and easy to maintain.

客户信息管理Client Information Mgt 客户专属价维护Exclusive Price Mgt 默认价格表维护Default Price List Mgt 业务规则沉淀Business Rules Accumulation

💰 欠账跟踪与账单流水管理Debt Tracking & Billing Flow Management

除了开单流程，系统还支持客户欠账总览、手动记账、回款登记和账单流水维护。

订单生成后也可以累计到账单中，把“开单”与“后续收款/欠款跟进”连接起来，让门店经营数据不再分散在不同工具里。

Beyond ordering, the system supports a customer debt overview, manual bookkeeping, payment registration, and billing flow maintenance.

Generated orders can accumulate into bills, bridging "order creation" with "subsequent payment/debt follow-ups", ensuring business data isn't scattered across multiple tools.

欠账总览Debt Overview 回款登记Payment Registration 账单流水维护Billing Flow Mgt 开单与记账联动Order & Billing Integration

☁️ 登录与云端共享协作Login & Cloud Shared Collaboration

系统支持店员登录，并将客户、价格、历史单据和账单数据接入云端共享。

这样多个店员可以基于同一套业务数据协作，而不是各自维护分散记录，也为后续系统稳定化和多人使用场景打下基础。

The system supports staff logins and connects clients, prices, historical documents, and billing data to cloud sharing.

This allows multiple clerks to collaborate based on the same business data instead of maintaining scattered records individually, laying the foundation for system stabilization and multi-user scenarios.

店员登录Staff Login 云端共享数据Cloud Shared Data 多人协作Multi-user Collab 统一业务数据源Unified Business Data

技术架构 Architecture Diagram Architecture Diagram

表现层 Presentation LayerPresentation Layer

承载登录、录单、预览、历史、数据库管理和账单等页面 Hosts Login, Order Entry, Preview, History, DB Management, and Billing pages

登录页 LoginPageLoginPage 录单页 OrderEditorPageOrderEditorPage 预览页 PreviewPagePreviewPage 历史记录 HistoryPageHistoryPage 数据库管理 DatabaseManagerPageDatabaseManagerPage 账单中心 BillingPageBillingPage

业务层 Application LayerApplication Layer

处理订单解析、价格匹配、校验、预览生成、历史重生成和账单联动等前端核心逻辑 Handles frontend core logic: parsing, price matching, validation, generation, and billing sync

本地规则解析 Local ParsingRule-based Local Parsing 客户资料自动补全 Profile Auto-fillProfile Auto-fill 价格智能匹配 Price MatchingPrice Matching 表单校验 Form ValidationForm Validation 销货单预览生成 Preview GenerationPreview Generation 账单同步触发 Billing SyncBilling Sync Trigger 历史记录重生成流 History RegenerationHistory Regeneration

本地状态层 Local State LayerLocal State Layer

负责本地数据库回退、编辑态保留、预览缓存和异常兜底保障业务连贯性 Responsible for local DB fallback, editor snapshots, cache, and error boundaries for continuity

本地业务数据库 Local DBLocal DB Fallback 本地历史单据 Local HistoryLocal History 编辑态快照 Editor SnapshotEditor Snapshot 预览缓存 Preview CachePreview Cache 挂起的历史状态 Pending StatePending State

服务层 Service LayerService Layer

连接云端认证、共享业务数据和自定义轻量登录接口 Connects to cloud authentication, shared business data, and lightweight custom login APIs

/api/login-accounts /api/staff-login Supabase Auth Supabase Database

数据层 Data LayerData Layer

统一存储客户、价格、销货单和账单等核心业务数据表 Unified storage for core tables: clients, prices, invoices, and billing entries

profiles customers customer_prices default_prices invoices invoice_items billing_entries

架构设计亮点 Architecture Highlights Architecture Highlights

前端优先设计：以前端优先的方式设计主流程，保证录单和开单操作足够轻量、直接。
云端一元化管理：使用共享云端数据库统一管理多门店终端的客户、价格、销货单和账单数据。
本地容灾与体验兜底：通过本地回退和状态保留机制，降低云端网络不稳定对业务流程造成的阻断风险。
定制化轻服务扩展：用轻量 Serverless API 补足店员登录场景，让通用系统认证更贴近真实门店的使用心智。

Frontend-First Design: Designed the main flow with a frontend-first approach to ensure order entry and processing are lightweight and direct.
Unified Cloud Management: Utilizes a shared cloud database to uniformly manage client, pricing, invoice, and billing data across multiple store terminals.
Local Fallback & Reliability: Mitigates the risk of business disruption caused by unstable cloud networks through local fallback and state retention mechanisms.
Custom Lightweight Services: Supplemented staff login scenarios with lightweight Serverless APIs, aligning universal authentication systems closer to real-store mindsets.

技术栈与核心决策 Tech Stack & Decisions Tech Stack & Core Decisions

Tech Stack

React 18TypeScriptViteReact RouterSupabaseVercel Serverless APIhtml-to-image

核心决策 1：在引入高成本的 OCR / LLM 之前，优先采用轻量且稳定的基于规则本地解析 (Rule-based local parsing)，优先跑通最小可行性流程。
核心决策 2：主流程核心单据优先保存并进入预览，次要数据利用后台进行异步同步 (Core invoice save first, secondary sync later)，以此换取极致的交互响应速度。

Core Decision 1: Before introducing costly OCR/LLM APIs, prioritized a lightweight and stable Rule-based local parsing approach to validate the MVP flow first.
Core Decision 2: Designed the main flow to save core invoices first and jump to previews, while secondary data triggers asynchronous sync in the background, trading sync perfection for ultimate interactive response speed.

产品演进路线 Product Roadmap Product Roadmap

2026-03-14 · 项目启动Mar 14, 2026 · Project Kickoff

发现真实痛点，需求抽象Discover Pain Points, Abstract Requirements

基于对家庭门店业务场景的长期观察，提炼“效率瓶颈”，决定通过 Vibe Coding 模式快速介入验证。 Based on long-term observation of family store business scenarios, extracted the "efficiency bottleneck" and decided to quickly validate via Vibe Coding.

2026-03-15 · MVP 闭环Mar 15, 2026 · MVP Closed-loop

前端主链路跑通Frontend Main Link Established

实现“文本解析输入 → 动态渲染预览页”的最短闭环，确认业务逻辑的程序化可行性。 Achieved the shortest closed-loop of "Text Parsing Input → Dynamic Preview Rendering", confirming the programmatic feasibility of business logic.

2026-03-16 – 03-18 · 体验与功能迭代Mar 16-18, 2026 · UX & Feature Iteration

UX 护航与复用扩展UX Safeguard & Reusability Expansion

实装 PNG 导出能力；通过 SessionStorage 解决切页数据丢失问题；增加历史记录回填编辑机制。 Implemented PNG export; solved page-switching data loss via SessionStorage; added history backfill editing mechanism.

2026-03-23 · Phase 3 架构升级Mar 23, 2026 · Phase 3 Architecture Upgrade

拥抱云端多用户协同Embracing Cloud Multi-user Collab

全面接入 Supabase，落地店员登录体系与 RLS 数据隔离；补齐账单管理闭环，系统进入生产可用态。 Fully integrated Supabase, landed staff login systems and RLS data isolation; completed billing management loop, entering a production-ready state.

2026-03-26 – 03-27 · 性能调优与部署上线Mar 26-27, 2026 · Tuning & Deployment

异步解耦与持续交付安全Async Decoupling & Continuous Delivery

拆分主链路，采用“先保存核心单据并反馈预览，后台异步同步客户与账单”的策略；完成 Vercel 安全部署。 Decoupled the main link by adopting a "save core invoices & preview first, background async sync later" strategy; completed secure deployment on Vercel.

2026-03-28+ · 真实验证Mar 28+ 2026 · Real-world Validation

门店实测上线Store Beta Launch

交由真实店员操作使用，持续收集来自一线的工作流反馈并开启后续迭代。 Handed over to real clerks for operation, continuously collecting frontline workflow feedback for future iterations.

验证反馈与迭代 Validation & Iteration Validation & Iteration

面对真实操作环境，我根据反馈结果进行了以下核心逻辑的复盘与修缮：

Facing the real operating environment, I reviewed and repaired the following core logics based on feedback:

已落地的体验修缮Landed UX Repairs

顺应心智的账单逻辑：实测发现，若客户当前已有欠账记录，生成新单时“挂账”往往是必选项。系统现已优化为「自动嗅探历史并默认勾选累计账单」，降低了漏点概率。
原生分享突破限制：针对移动端浏览器长按“复制”图片失效的局限性，追加调用 Native `分享` API，可直跳微信选择好友，链路更无缝。
超长单据动态伸缩：起初只支持固定 8 行的单据模板。针对大宗采购，重构了底层逻辑：超 8 件商品时，自动开启纵向伸缩渲染为「长图」，打破规格边界。

Mindset-driven Billing Logic: Testing revealed that if a client has existing debt, "putting it on the tab" for new orders is usually mandatory. The system now "auto-sniffs history and defaults to accumulating the bill", lowering omission risks.
Breaking Native Sharing Limits: Addressing the limitation of mobile browsers failing on long-press "copy image", added Native Web Share API calls to directly jump to WeChat friends list, making the flow seamless.
Dynamic Scaling for Long Receipts: Initially supported a fixed 8-row template. Refactored for bulk purchases: when exceeding 8 items, it auto-enables vertical scaling to render as a "long image", breaking dimension limits.

权衡中与 Next PhaseTrade-offs & Next Phase

解析引擎的成本权衡：现阶段高度依赖“正则结构识别”，若遇到输入全角符号或顺序倒置会导致解析失准。方案碰撞：是在现行正则基础上持续打补丁？还是接入真实 LLM API（如 DeepSeek/Kimi）进行泛化解析？目前正与店长测算 ROI。
UX 文案的误解修复：旧版本中的“返回编辑”导致用户以为能直接覆盖当前账单（实际是开启全新副本）。即将把文案拆分为动作导向更强、无歧义的“再开一单”与“修改当前单”。
国内云基建迁移：考虑将整体数据迁移至腾讯云 CloudBase，规避海外 BaaS 访问延迟。

Parsing Engine Cost Trade-offs: Currently relies heavily on Regex structure recognition, failing on full-width symbols or inverted order inputs. Dilemma: Keep patching Regex, or integrate a real LLM API (like DeepSeek/Kimi) for generalized parsing? Currently calculating ROI with the owner.
UX Copy Misunderstanding: The old "Return to Edit" led users to think it overwrites the current bill (it actually opens a new copy). Planning to split the copy into action-oriented "Create Another Order" and "Modify Current".
Domestic Cloud Migration: Considering migrating all data to Tencent CloudBase to avoid overseas BaaS access latency in China.

阶段性业务成果 Key MetricsKey Metrics

<1 min

已入库客户平均开单耗时Avg. order time for logged clients

原流程需 2–5 分钟Prev. workflow took 2-5 mins

150+

真实门店有效订单Valid orders in real stores

提效 500%+ 目标达成Efficiency +500% goal reached

2,000+

数字化管理的客户档案Digitally managed client profiles

含专属定价与全量欠账Incl. exclusive prices & full debts

餐饮业省人化实体产品设计 Labor-Saving Physical Product Design for F&B

Engineering Design Project (EDP) · Tokyo Tech × Dassault Systèmes

6人跨专业团队 (Team Project)6-Member Interdisciplinary Team 2025.09 – 2026.02Sep 2025 – Feb 2026 核心职责：跨文化沟通枢纽 · 用户调研主导 · 痛点挖掘Core: Cross-cultural bridge · Led User Research · Pain-point Mining

跨文化沟通枢纽 Cross-cultural Bridge Cross-cultural Bridge

依托 8 年的留日经验与中、日、英三语沟通能力，我与 6 人跨专业团队（融合理工、机械、美术设计等）与日本企业（ダッソー・システムズ Dassault Systèmes）进行深度对接，更顺利主导了部分深度采访与最终的公开日语发表。

Relying on 8 years of study experience in Japan and Trilingual (CN, JP, EN) communication skills, I collaborated deeply with a 6-member interdisciplinary team (engineering, mechanics, art design, etc.) and the Japanese enterprise (Dassault Systèmes), successfully leading in-depth interviews and the final public presentation in Japanese.

命题方向 Problem StatementProblem Statement

企业命题Corporate Prompt

设计一款能进一步推动餐饮业省力化、省人化的产品

Design a product that further promotes labor-saving and manpower reduction in the food and beverage industry.

— Dassault Systemes 命题 (課題テーマ: 飲食業におけるさらなる省力化・省人化を進めるプロダクトをデザインせよ)

深度用户调研 User ResearchIn-depth User Research

场次用户深度访谈In-depth User Interviews

家门店实地工作流调研On-site Workflow Studies

种不同餐饮业态覆盖F&B Formats Covered

名跨专业背景成员协同Interdisciplinary Members

访谈矩阵立体：涵盖了一线店员、店长、消费者，乃至餐饮器械技术人员与工业产品设计师。

Comprehensive Matrix: Covered frontline staff, managers, consumers, F&B equipment technicians, and industrial designers.

调研范围广泛：从英国风 Pub、家庭餐厅 (ファミレス)、居酒屋，最终战略聚焦至高频流水作业的连锁咖啡店。

Broad Scope: From British Pubs, Family Restaurants (ファミレス), Izakayas, eventually focusing strategically on high-frequency workflow chain coffee shops.

现场课题与痛点分析 On-site Pain Points On-site Pain Points

🚨 高频且繁重的补货作业High-Frequency Heavy Restocking

通过实地调研，我们发现大型连锁咖啡店（如星巴克）的员工面临着巨大的后厨工作负荷：最短每 15 分钟，就需要将一整箱（约 12kg）的牛奶搬运并补充进冷柜中。

这种纯依赖托盘搬运和人手操作的高频重复动作，极其消耗员工体力与接客时间。

Through field research, we discovered that employees in large chain coffee shops (like Starbucks) face immense back-of-house workloads: As frequently as every 15 minutes, they must transport and restock a full box (approx. 12kg) of milk into the refrigerator.

This highly repetitive action, relying purely on tray carrying and manual handling, severely exhausts staff stamina and customer service time.

高频作业 (15分钟/次)High Frequency (15min/time) 体力负担重 (12kg/箱)Heavy Burden (12kg/box)

🧊 结构受限导致的 FIFO 积压困境FIFO Backlog due to Structural Limits

相较于便利店双向开门的饮料柜（后进前出），咖啡店冷柜通常仅有一个开门（单向进出）。这导致新补充的牛奶很容易直接堵在最前面，旧牛奶则被不断推积在深处。

为了严格遵守 FIFO（先进先出）原则，员工每次补货都必须手动将旧牛奶移出、确认保质期，再把新牛奶塞入后方。如果为了省事忽略顺序，又会造成极大的过期报废损耗。

Unlike convenience store beverage coolers with two-way doors (back-load, front-take), coffee shop refrigerators usually only have one door (one-way access). This causes newly restocked milk to easily block the front, pushing old milk deep inside.

To strictly adhere to FIFO (First-In, First-Out) principles, staff must manually pull out old milk, check dates, and stuff new milk to the back during every restock. Skipping this causes massive expiration waste.

单向进出的物理结构限制One-way Physical Structure 极易形成隐性过期积压死角Hidden Expiration Blind Spots 高昂的商品报废损耗成本High Waste Costs

目标用户画像 User PersonaUser Persona

☕

连锁咖啡店员（如星巴克等）Chain Coffee Clerk (e.g., Starbucks)

20 代 · 兼职或正式员工 · 需兼顾前台接客与后台库存管理 20s · Part-time/Full-time · Must balance frontline service & backend inventory

日常任务Daily Tasks

接客、饮品制作、库存整理、牛奶补货、FIFO 排序，全在繁忙时段的间隙极限压缩完成。

Greeting, drink making, inventory sorting, milk restocking, FIFO organizing, all compressed into gaps during rush hours.

核心痛点Core Pain Points

牛奶补货需逐个翻看保质期并进行空间挪腾，步骤繁杂，稍微分心就容易排错顺序。

Restocking milk requires checking dates one by one and shifting space. Complex steps make it easy to mess up the order if distracted.

目标Goals

希望补货动作达到“无脑化”，干掉人为思考顺序的环节，把时间还给前台服务。

Wants restocking to be "brainless", eliminating the need to manually think about order, giving time back to customer service.

流程重塑：从人工到产品干预 User JourneyJourney Redesign: Manual to Product Intervention

⚠ Before — 极其消耗心智的人工 FIFO 排序⚠ Before — Mentally Exhausting Manual FIFO

新牛奶
入库New Milk
Arrives

逐个查看
保质期Check Dates
One-by-One

需逐瓶
翻看确认日期Flip each
to check

手动腾挪
排列位置Manually
Shift Space

取出旧货
让出后排空间Pull old stock
make room

复原旧货
前置Restore Old
to Front

严重依赖
瞬时记忆执行Relies heavily
on memory

跑回前台
接客Run Back
to Front

大量浪费
黄金接客时间Wastes golden
service time

✓ After — 使用本物理装置辅助后的顺滑流程✓ After — Smooth Flow with Our Physical Device

新牛奶
入库New Milk
Arrives

直接放入
专用装置Drop into
Device

无脑丢入即可Just drop it in

装置物理
自动导向Auto Physical
Routing

临期商品必前置Older stock front

立即返回
前台接客Return Instantly
to Front

专注客户微笑Focus on smiles

产品运作原理 How to Use: Stock RideHow to Use: Stock Ride

⚙️ 极简的两步操作体验Minimalist 2-Step Experience

我们将繁杂的日期对比和空间挪腾工作，全部交给了精巧的物理结构。

使用本装置（命名为 Stock Ride），店员仅需闭眼执行两个动作：

We delegated all the complex date comparisons and spatial shifting to a clever physical structure.

Using this device (named Stock Ride), clerks literally only perform two blind actions:

补充 (IN)：将新拿到的牛奶包直接从装置上方放入。
抽取 (OUT)：需要使用时，直接从装置最下方抽取。
Restock (IN): Drop the newly received milk bags directly from the top of the device.
Extract (OUT): When needed, just pull from the very bottom of the device.

通过力学轨道的巧妙引导，只要重复这简单的 ①+② 步，物理装置就会自动完成完美的先入先出（FIFO）闭环，彻底释放一线员工的心智负担！

Through the clever guidance of mechanical tracks, repeating steps ①+② ensures the physical device automatically completes a perfect First-In-First-Out (FIFO) loop, totally freeing frontline workers' mental load!

用户故事 User StoriesUser Stories

用户角色 (As a...)User Role (As a...)	需求描述 (I want to...)Requirement (I want to...)	核心价值 (So that...)Value (So that...)
一线店员Frontline Clerk	在往冰柜补货时，不需要人工判断日期，塞进去就能自动完成先进先出排列。	When restocking fridges, no manual date checks are needed; just dropping it in auto-arranges FIFO.	彻底消灭排错顺序的可能，不再需要背诵复杂的操作规范。	Completely eliminates sequence errors, no more memorizing complex ops manuals.
前台服务生Server	能够把“取奶”和“补奶”的时间压缩到极致。	Able to compress the time for "taking" and "restocking" milk to the absolute minimum.	能随时响应前台顾客的需求，提升门店翻台率和好评度。	Can respond to front-desk customer needs instantly, boosting table turnover and ratings.
门店店长Store Manager	即便新入职的打工兼职（临时工）也能不打折扣地执行 FIFO 标准。	Even newly hired part-timers (temps) can execute FIFO standards flawlessly.	避免高昂的商品报废损耗成本，提升门店整体利润率。	Avoids high goods expiration waste costs, improving overall store profit margins.

产品宣传与概念展示 Promo VideoPromo Video

除了硬件本体，另为产品策划并制作了一支全方位的产品商业推介视频。该宣传视频的内容策划、分镜脚本设计、实地素材拍摄以及最终的剪辑制作均由我个人主导完成。

Beyond the hardware, we also produced a comprehensive commercial promo video. The content planning, storyboard design, field shooting, and final editing of this video were entirely led and completed by me.

阅读 Blog (Medium) ↗Read Blog (Medium) ↗

影响力验证 Impact & ValidationImpact & Validation

背景高度融合的专业团队Highly diverse professional team

涵盖：信息通信、融合理工、机械、环境设计与美术Incl: Info Comm, Transdisciplinary Eng, Mechanics, Art Design

300+

线下发表现场听众规模Offline presentation attendees

同步开启无限制免费全网线上推介直播Simultaneous free online live broadcast

30+

企业评委与高阶行业专家Corporate judges & industry experts

汇聚 5 家联合命题企业及 10 余家投资关注方Gathered 5 joint sponsor companies & 10+ investors

AI 故事转漫画互动生成系统 AI Story-to-Comic Generator System

工大祭 2025 参展项目 (AI Story-to-Comic Generator) Koudaisai 2025 Exhibition Project

展览策划与开发项目Exhibition Planning & Dev 2025.10 – 11Oct-Nov 2025 我的角色：LLM 工具开发 · Prompt 流程设计 · 现场运营支撑Role: LLM Tool Dev · Prompt Flow Design · On-site Ops

项目背景与展览概念 Exhibition Concept Exhibition Concept

活动背景：Background: 本项目为 2025 年东京科学大学学园祭（工大祭）中的线下互动体验展位。 This project was an offline interactive exhibition booth at the 2025 Tokyo Institute of Science Campus Festival (Koudaisai).

核心概念：Core Concept: 打造一个“人人皆可成为创作者”的 AI 漫画交互站。参观者只需向工作人员简单口述或分享一段自己的日常趣事、奇思妙想，系统便能实时将其转化为一张独一无二的专属日系漫画，并同步展示在会场的 Miro 数字墙上。 Creating an AI comic interactive station where "everyone can be a creator". Visitors simply dictate or share a daily anecdote or idea with the staff, and the system instantly transforms it into a unique, exclusive Japanese-style comic, simultaneously displayed on the venue's digital Miro wall.

项目定位：Project Positioning: 这是一次面向大众的 AIGC（生成式 AI）零距离科普体验，同时调研一般民众（尤其是老人和小孩）在面对 AI 创作时关注的修改点、心理接受度及倾向偏好等，为后续的学术科研打下了坚实的数据与洞察基础。 This was a close-up AIGC (Generative AI) educational experience for the general public, while also researching the general public's (especially elders and children) psychological acceptance, modification concerns, and preferences when facing AI creation, laying a solid data foundation for future academic research.

痛点与解法 Problem & Solution Problem & Solution

现场瓶颈On-site Bottleneck

在展览中，面对观众滔滔不绝的口述故事，工作人员需要一边听一边手动打字提取要素，且每次都要重复输入极其冗长的绘画参数（Prompt），不仅耗时巨长，还非常容易漏掉核心情绪，导致大排长龙。

Facing visitors' endless spoken stories, staff had to listen and manually type to extract elements simultaneously, repeatedly entering extremely lengthy drawing prompts. This was hugely time-consuming, prone to missing core emotions, and caused massive queues.

AI 破局AI Breakthrough

我主动提议并主导开发了一个基于本地部署的 LLM 辅助流转工具：实现观众故事的结构化转录、自动化特征关键词提取，并瞬间拼装为专业的漫画提示词。把人从“打字员”解放为“体验向导”。

I proposed and led the development of a locally deployed LLM assistance tool: achieving structured transcription of visitor stories, automated feature keyword extraction, and instant assembly into professional comic prompts. Freeing humans from being "typists" to becoming "experience guides".

系统架构与管线 System Pipeline System Pipeline

故事捕捉源头Story Capture Source

工作人员引导参观者分享经历，口语化的叙事被捕获。Staff guide visitors to share experiences; colloquial narratives are captured.

Ollama 本地 LLM 引擎降噪过滤Ollama Local LLM Noise Filtering

部署在本地的离线大模型对输入信息进行结构化清洗，规避展会糟糕的公网断连风险。Locally deployed offline LLM cleans structural input, avoiding terrible public network disconnection risks at exhibitions.

绘画特征元数据提取Drawing Feature Metadata Extraction

精准抓取：视觉场景、主角特征、核心动作与情感基调。Accurate capture of: visual scenes, protagonist features, core actions, and emotional tone.

Prompt 组装车间Prompt Assembly Factory

将特征数据映射拼装为 Gemini 可识别的标准指令。Mapping feature data into standard instructions recognizable by Gemini.

Gemini 终端生图与反馈调整Gemini Generation & Feedback Adjustment

（依据用户喜好微调风格指令，完成漫画渲染）(Fine-tune style commands based on user preference to complete comic rendering)

Miro 画板公开展示所有漫画Miro Board Public Display

共 88 张专属漫画，现场展览 + 问卷收集反馈Total 88 exclusive comics, on-site exhibition + questionnaire feedback

系统操作界面 UI & InteractionUI & Interaction

工程亮点 Technical HighlightsTechnical Highlights

🤖

离线大模型驱动力Offline LLM Drive

通过 Ollama 在展位笔电上硬核拉起本地大模型引擎，彻底斩断对现场恶劣公网及昂贵 API 的依赖。

Spun up a local LLM engine via Ollama directly on the booth's laptop, completely cutting dependency on terrible on-site public networks and expensive APIs.

🖥️

闭环可交互原型Closed-loop Prototype

使用 Python/Node.js 构建了约 20 个文件的中小型独立应用，包含表单、动作响应与数据承载的完整骨架。

Built a medium-sized standalone app with Python/Node.js (~20 files), featuring forms, action responses, and full data-carrying skeletons.

⚡

高阶 Prompt 流水线Advanced Prompt Pipeline

设计了一套稳定的多层级推理链（思维链）：让模型依次扮演聆听者、提炼者和专业绘画指导三个身份。

Designed a stable multi-level Chain of Thought (CoT): letting the model sequentially act as a listener, extractor, and professional art director.

活动势能与回响 Event ImpactEvent Impact

250+

两日间驻足展区人数2-day booth visitors

100+

重度交互与体验用户Heavy interactive users

高质量故事插画展示High-quality story comics

30+

有效定性体验洞察问卷Qualitative UX insights

张紫琼的作品集 Zhang Ziqiong's Portfolio

我的项目 My Projects My Projects

门店智能开单与账单管理系统 Smart Order & Billing Management System

📝 智能录单与订单编辑Smart Order Entry & Editing

🧾 销货单生成与图片化导出Receipt Generation & Image Export

⏳ 历史记录管理与重生成History Management & Regeneration

🗂️ 客户资料与价格规则管理Client Profiles & Pricing Rules

💰 欠账跟踪与账单流水管理Debt Tracking & Billing Flow Management

☁️ 登录与云端共享协作Login & Cloud Shared Collaboration

餐饮业省人化实体产品设计 Labor-Saving Physical Product Design for F&B

🚨 高频且繁重的补货作业High-Frequency Heavy Restocking

🧊 结构受限导致的 FIFO 积压困境FIFO Backlog due to Structural Limits

⚙️ 极简的两步操作体验Minimalist 2-Step Experience

AI 故事转漫画互动生成系统 AI Story-to-Comic Generator System

学术研究 Academic Research Academic Research

教育背景 Education & Background Education & Background