广州中望龙腾软件股份有限公司 (ZWSOFT) — 软件框架小组 | 软件开发实习生

2026.07 - 至今

工作地点：上海。就职于共性技术研究院 / 共性组件研发部 / 综合软件技术组 / 软件框架小组，参与 ZW3D（中望 3D CAD/CAM，千万行级 C++ 代码库）模块化重构项目，负责代码质量分析平台的主开发工作（Flask + 浏览器原生 SVG/ECharts/Panzoom.js），覆盖全局变量总览、模块依赖矩阵、可视化依赖图与历史趋势对比四大模块；同步参与 Clang 静态分析工具链中头文件与全局变量扫描工具的迭代维护。

• 工作内容整理中，敬请期待。

陕西天权时空智能科技有限公司 / 陕西航天宏图体系 — 空天产品研发部 | C++ 开发实习生

2026.04 - 2026.06

参与航空航天及部分军工涉密内网项目的研发与工程交付，覆盖 C++/Qt 开发、跨平台编译适配（银河麒麟 / 统信 UOS / Ubuntu / Windows / ARM 嵌入式）、Docker 打包部署与 AI 工程化落地。

• 地面-卫星通信模拟（天权星控系统核心模块，C++ / Qt）：主动定位并修复通信模拟模块缺陷，完成 Windows 与 Ubuntu 双平台打包；识别部署目标为 ARM 架构嵌入式 Linux 后重新配置编译环境完成适配，部署至正点原子开发板并基于模拟数据完成验证。
• 星控算法测试（天权星控系统）：负责测试服务 Docker 镜像构建与启动验证，完善前端页面，修复 Python 后端缺陷并补充功能，完成功能优化、测试验证与打包部署。
• 跨平台编译与集成适配：在银河麒麟 V10 从 Qt 源码编译出完整 Qt 环境，排查解决系统权限与环境问题；完成 6GB+ 军工涉密系统中多个 Qt 插件的编译修复，将上千条写死的绝对路径重构为相对路径，处理 qmake 报错与依赖缺失；在统信 UOS 完成 UE 源码编译与 Cesium for Unreal 插件集成，打通 Windows 交叉编译 Linux 版本流程；完成 AFSIM 军事仿真框架编译打包、Docker 环境适配与文档沉淀，将该项编译交付由历史的 1 天以上压缩至约 3–4 小时。
• 内网文件全流程管控系统（独立交付）：基于 Claude Code + Hermes Agent，从需求拆解、系统设计、功能开发到部署交付与交接文档全程独立完成，系统已正式落地用于公司内网；并与高级工程师共同探讨结合交换机实现内网“实名”标定上网的管控方案。
• AI 工程化专项：将公司本地部署的 Qwen 大模型接入 OpenCode 与 Hermes Agent 并完成多机部署测试；基于错误日志与多轮对照实验，定位模型“标称 256K、实测仅约 16K”的上下文配置异常并推动修复；基于 Kanban 搭建项目管理员 / 开发 / 测试多 Profile 的多 Agent 协作流程并输出使用案例；基于 Hermes API + CopilotKit 完成生成式 UI PoC。
• 涉密现场独立交付：先在资深工程师远程指导下完成地磁应用系统多组件（Docker / Nginx / MySQL / PostgreSQL / RabbitMQ / Redis）环境部署与调试；后独自赴涉密单位现场，在国防麒麟系统上完成大型平台子系统的部署交付，现场处置安装介质文件缺失等突发问题，在同步进行培训/演示的环境下于 2–3 小时内完成主体部署。
• AI 辅助跨栈交付与团队赋能：依据 1400+ 页检测报告处理外包 Java 项目（606 个文件 / 7.2 万行代码），结合 Git 分批备份与回滚工作流，将问题数由 4076 降至 100 以内；为 Qt 项目补写 1 万+ 行注释，注释率由约 13% 提升至约 27%；面向全公司开展约 80 分钟 AI 工具使用分享，并为运营总监、总经理助理及多位运维同事完成 AI 工具安装配置与讲解。

Oh-My-OpenLog

基于多设计模式实现的高性能 C++ 日志系统，支持多级别、多落地方向及同步/异步输出。采用双缓冲区生产消费模型，测试表明多线程异步非安全模式每秒可处理超 93 万条日志，有效解决磁盘 IO 阻塞问题。

C++11, 设计模式, 多线程, 双缓冲区

C++ RPC 组件库设计

基于 Muduo 的轻量级分布式 JSON-RPC 框架，采用抽象层、具象层与业务层三层架构设计。实现了基于 LV 格式的应用层协议解决粘包问题，内置动态服务发现与注册机制，并集成发布订阅(Pub/Sub)模型。

C++11, JsonCpp, Muduo, Linux, 分布式

高并发内存池 (Memory Pool)

基于 tcmalloc 核心架构实现的高并发内存池，解决多线程下的锁竞争与内存碎片问题。采用 Thread Cache、Central Cache 与 Page Cache 三级缓存架构，实现线程局部无锁分配，显著提升高并发场景下内存分配效率。

C++, 操作系统内存管理, 哈希桶, 多线程并发

ZWSOFT · Shanghai — Software Framework Group | Software Development Intern

Jul 2026 - Present

Based in Shanghai. Common Technology Research Institute / Common Components R&D Dept. / General Software Technology Group / Software Framework Group. Joined the modular refactoring of ZW3D (ZWSOFT 3D CAD/CAM, a 10M+-line C++ codebase) as the main developer of its code quality analysis platform (Flask + native-browser SVG/ECharts/Panzoom.js), covering four modules: global variables overview, module dependency matrix, visual dependency graph, and historical trend comparison; also maintaining the header-file and global-variable scanners in the Clang static analysis toolchain.

• Details coming soon.

Shaanxi Tianquan Shikong Intelligent Tech / PIESAT — Aerospace Products R&D | C++ Engineering Intern

Apr 2026 - Jun 2026

Participated in the R&D and engineering delivery of aerospace and classified defense internal-network projects, covering C++/Qt development, cross-platform builds (Kylin V10, UOS, Ubuntu, Windows, ARM embedded), Docker packaging/deployment, and AI engineering practices.

• Ground-to-satellite communication simulation (core module of the Tianquan Satellite Control System, C++ / Qt): proactively located and fixed defects in the communication simulation module and delivered packages for both Windows and Ubuntu; identified the deployment target as ARM embedded Linux, reconfigured the build environment accordingly, deployed to an Atom (正点原子) dev board, and validated against simulated data.
• Satellite control algorithm testing (Tianquan Satellite Control System): built and verified Docker images for the test service, improved the front-end pages, fixed defects and added features in the Python backend, and completed feature optimization, test verification, packaging, and deployment.
• Cross-platform compilation and integration: built a complete Qt environment from source on Kylin V10 while resolving permission and environment issues; repaired the builds of multiple Qt plugins in a 6GB+ classified defense system, refactoring thousands of hardcoded absolute paths to relative paths and fixing qmake errors and missing dependencies; compiled UE from source and integrated the Cesium for Unreal plugin on UOS, enabling the Windows-to-Linux cross-compilation pipeline; completed AFSIM military simulation framework builds, Docker adaptation, and documentation, cutting that build delivery from 1+ day historically to about 3-4 hours.
• Internal File Management System (independently delivered): built end-to-end with Claude Code + Hermes Agent — from requirements breakdown, system design, and feature development to deployment and hand-off documentation; now officially in production on the company intranet; also discussed with senior engineers a switch-based solution for real-name traffic control on the intranet.
• AI engineering initiatives: integrated the company's locally deployed Qwen LLM into OpenCode and Hermes Agent with multi-machine deployment testing; via error logs and controlled experiments, pinpointed a context configuration bug (nominal 256K vs. ~16K actual) and drove the fix; built a multi-Agent collaboration workflow with separate admin/dev/test profiles on Kanban and documented use cases; delivered a generative UI PoC with Hermes API + CopilotKit.
• On-site classified delivery: first deployed and debugged a multi-component geomagnetic application environment (Docker / Nginx / MySQL / PostgreSQL / RabbitMQ / Redis) under a senior engineer's remote guidance; then traveled alone to a classified site to deliver a major platform subsystem on National Defense Kylin, handling on-site incidents such as missing installation media and completing the core deployment within 2-3 hours while training/demo sessions ran in parallel.
• AI-assisted cross-stack delivery and team enablement: worked through an outsourced Java project (606 files / 72,000 lines) against a 1400+ page inspection report using a Git batch backup/rollback workflow, cutting issues from 4076 to under 100; added 10,000+ lines of comments to a Qt project, raising comment coverage from about 13% to 27%; delivered an 80-minute company-wide AI tooling workshop and configured/coached the operations director, assistant to the GM, and several ops colleagues on AI tools.

Oh-My-OpenLog

A high-performance C++ logging system supporting multi-level, multi-destination, and async/sync outputs. Utilizes a double-buffering producer-consumer model, achieving over 930,000 logs/sec in async non-safe mode.

C++11, Design Patterns, Multithreading, Double Buffering

JsonRpc Framework

A lightweight distributed JSON-RPC framework based on Muduo, implementing a 3-tier architecture. Resolves TCP sticky packets using LV format, features dynamic service discovery, and integrates a Pub/Sub event model.

C++11, JsonCpp, Muduo, Linux, Distributed

High-Concurrency Memory Pool

A high-concurrency memory allocator modeled after Google's tcmalloc. Uses a 3-tier cache (Thread, Central, Page) for lock-free thread-local allocation, minimizing lock contention and memory fragmentation.

C++, OS Memory Mgmt, Hash Buckets, Concurrency