A native macOS menu-bar productivity utility featuring region capture & annotation, scrolling capture, screen recording, OCR, selection/screenshot translation, floating pins, clipboard history, and unified history. Built with Swift 6, SwiftUI + AppKit, ScreenCaptureKit, Vision, and Translation with an on-device-first architecture and optional cloud OCR/translation.
Next-generation enterprise AI Agent platform frontend built atop an in-house AI SDK to unify agent conversations, long-running task execution, tool calls, skill extensions, memory, and workspace files. Owned the chat runtime, MCP tool management, skill hub, memory & workspace views, attachment uploads, and multi-format file previews supporting streaming, runtime states, human-in-the-loop approval, and permission modes.
Enterprise AI application workbench frontend offering low-code visual agent creation, workflow orchestration, plugin/MCP integration, knowledge bases, AI PPT generation, debugging, review, publishing, and marketplace distribution. Led AI PPT integration, complex workflow node extensions, plugin/MCP parameter refactoring, and release review flows.
Dual city-commute WeChat mini-program matrix built from 0 to 1—Traffic Restriction Helper for non-local drivers in Guangzhou/Shenzhen and E-Bike Pass for e-bike riders—serving ~450k cumulative users. Built with Taro 4 + React sharing one Spring Boot backend, featuring multi-appType config, map overlays, real-time camera alerts, crowd-sourced moderation, COS uploads, memberships, and ad-unlock flows.
Full-stack photography gear discovery, spec comparison, personal kit management, and AI gear advisor platform. Independently built the uni-app 3 + Vue 3 WeChat mini-program, Next.js 16 API & scheduled worker, a 13-agent Mastra AI advisory service, Vue 3 admin console, and Python scraping pipeline supporting multi-dimensional filtering, 44+ spec diff comparisons, gear locker management, and streaming context-aware AI recommendations.
High-density in-game quick-lookup WeChat mini-program for League of Legends Hextech ARAM—covering champion tiers, augments, augment-set synergies, builds, and skill orders. Built independently with uni-app + Vue 3, an Express query API, a Vue 3 admin portal, and Python Playwright pipelines ingesting normalized data from OP.GG, ARAM Mayhem, and official Tencent LOL sources.
Information system for monitoring and regulating Guangdong Province land assets, land supply, post-supply supervision, stock revitalization, and statistical reports. Owned complex forms, GIS map interactions, configurable list/form engines, domestic-stack migration refactoring, and first-paint optimization.
Contributed to e-invoicing workflows using Vue 2, qiankun, TDesign Vue, and Webpack—owning feature iterations, reusable tax UI components, and high-precision monetary and tax rate calculations.
Developed multi-end property contract signing flows with Vue 2, WeChat Mini-Programs, ant-design-vue, Vant, and Webpack—handling facial-recognition login, frontend RBAC, and cross-terminal layout compatibility.