曹劼成Cao Jiecheng

• 负责期货交易平台移动端端到端开发，覆盖 Flutter 前端、Swift 原生模块、Go 后端与 SQL 数据库；开发日内分时图与 MACD 指标组件，构建高性能 Canvas 图形库使首帧渲染低于 120 毫秒。Owned end-to-end development of the futures trading app, spanning the Flutter front end, Swift native modules, a Go backend and SQL storage. Built the intraday chart and MACD indicator components plus a high-performance Canvas graphics library that brought first-frame rendering under 120 ms.
• 完成 T 型期权报价表与自定义懒加载表格，配合 Go 后端 WebSocket 增量推送，滚动稳定 60 帧、内存占用降低约 30%；重构业务层与网络层提高代码复用率。Delivered the T-shaped option quote table and a custom lazily-loaded table fed by incremental WebSocket pushes from the Go server; scrolling held a steady 60 fps and memory use dropped about 30%. Refactored the business and network layers to raise code reuse.
• 定位并修复前后端类型转换崩溃问题、重构数据模型，消除同类崩溃。Diagnosed and fixed a front-end/back-end type-conversion crash by reworking the data model, eliminating that whole class of crash.
• 重构 Mock 数据工厂，统一管理模拟行情与空值、超大数据、网络超时等边界数据，扩大极端场景用例覆盖范围；经关键数据缓存与 IPv6 支持将冷启动缩短约 20%。Rebuilt the mock-data factory to centralise simulated market data and edge cases (nulls, oversized payloads, network timeouts), widening coverage of extreme scenarios. Cached key data and added IPv6 support to cut cold start by roughly 20%.

• 基于 JMeter 完善企业级接口自动化脚本库，将依赖 UI 客户端的手动触发改造为命令行加 Ant 脚本调度，集成 HTML 报告自动生成实现回归无人值守；补充用例使 API 覆盖率由约 70% 达到 90%。Extended the enterprise API automation suite in JMeter, replacing manual UI-client triggering with command-line plus Ant scheduling and automated HTML report generation to make regression runs unattended. Added cases that lifted API coverage from about 70% to 90%.
• 作为核心成员负责团队 UI 自动化平台业务层落地，运用 Python、Pytest、Selenium 封装 setUp/tearDown、等待策略与日志记录等测试组件，按 Page-Object 模式完成 60+ 页面元素定位封装。As a core member, built the business layer of the team's UI automation platform — Python, Pytest and Selenium components for setUp/tearDown, wait strategies and logging, with 60+ page elements encapsulated under the Page Object pattern.
• 将存量手工用例转写为自动化脚本，承担发布前冒烟与日常回归测试，采用自动化保底加人工验证复杂业务逻辑的组合模式，冒烟阶段缺陷发现率提高、回归周期缩短约 20%。Converted existing manual cases into automated scripts, owning pre-release smoke and daily regression runs. Combined automation as a floor with manual verification of complex business logic, raising defect discovery in smoke testing and cutting the regression cycle by about 20%.
• 基于 GitHub Actions 编写自动化触发与报告推送示例，为团队后续流水线迁移提供参考。Wrote a GitHub Actions example for automated triggering and report delivery as a reference for the team's pipeline migration.

面向预付卡发行、流通与兑换场景的链上管理系统，覆盖卡片发行、转账、销毁与权限分级等业务，卡片元数据采用去中心化存储保障长期可追溯。核心难点在于多资产额度的一致性维护与预付额度防超发，以及链上交易与前端余额展示的准实时同步，异常回滚与权限越权是重点质量关注方向。An on-chain management system for prepaid card issuance, circulation and redemption, covering card minting, transfers, burning and tiered permissions, with card metadata stored decentrally for long-term traceability. The hard parts were keeping multi-asset balances consistent and preventing over-issuance of prepaid credit, plus near-real-time sync between on-chain transactions and the balance shown in the UI. Rollback on failure and privilege escalation were the key quality concerns.

• 使用 Solidity 编写 ERC-1155 标准智能合约并部署至 Hardhat 本地链，覆盖卡片发行、转账、销毁及权限分级控制；采用 React 开发前端并集成 MetaMask 钱包登录，元数据与图片存储于 IPFS。Wrote ERC-1155 smart contracts in Solidity and deployed them to a local Hardhat chain, covering card issuance, transfer, burning and tiered permission control; built the front end in React with MetaMask wallet login, storing metadata and images on IPFS.

面向物体识别与自动分拣场景的嵌入式视觉小车，完成小车结构与六轴机械臂三维建模、视觉识别与运动控制的全链路实现，支持复杂环境下的自动导航与精准对接。核心难点在于受限算力下识别精度与实时性的平衡、视觉与雷达两类传感数据的融合稳定性，以及图像处理与运动控制并行时的时序可靠性。An embedded-vision robot for object recognition and automated sorting, spanning 3D modelling of the chassis and six-axis arm through vision recognition and motion control, with autonomous navigation and precise docking in cluttered environments. The hard parts were balancing recognition accuracy against real-time performance on constrained compute, keeping vision and ultrasonic sensor fusion stable, and preserving timing reliability when image processing and motion control run in parallel.

• 使用 SolidWorks 完成小车结构及六轴机械臂三维建模，输出 CAD 图纸与 3D 打印文件；在 OpenMV 平台训练颜色与物体识别模型，结合超声波雷达实现融合测距，建立 Kendryte K210 与电机控制板双向通信，以异步传输使图像处理与运动控制并行。Modelled the chassis and six-axis arm in SolidWorks and produced CAD drawings and 3D-print files; trained colour and object recognition models on the OpenMV platform, fused ultrasonic ranging for distance sensing, and set up bidirectional communication between the Kendryte K210 and the motor controller so image processing and motion control could run in parallel over asynchronous transfers.

面向校园资源共享场景的 Web 平台，支撑统一账号管理、资源存储、分类展示与多格式文件下载，并以积分机制激励资源共享行为。核心难点在于大文件断点续传的可靠性、全文检索的响应延迟控制，以及多用户并发下积分交易与资源数据的一致性，跨环境部署的安全性同样是关键约束。A web platform for sharing campus resources, supporting unified account management, resource storage, categorised browsing and multi-format downloads, with a points mechanism to encourage sharing. The hard parts were reliable resumable uploads for large files, keeping full-text search latency down, and preserving consistency between points transactions and resource data under concurrency. Secure cross-environment deployment was another key constraint.

• 使用 Django 与 MySQL 开发平台后端，设计并关联用户、文件、积分等核心数据表；完成断点续传、全文检索与积分交易功能，采用 Docker 部署并配置 Nginx 反向代理与 SSL 加密保障跨环境稳定。Built the backend with Django and MySQL, designing and relating the core tables for users, files and points; implemented resumable uploads, full-text search and points transactions, and deployed with Docker behind an Nginx reverse proxy with SSL to keep environments consistent.