The author developed the multiplayer drawing game Drawize for a Samsung Tizen OS app contest in late 2016, completing it in just 4 weeks, including a real-time multiplayer engine, but didn’t win an award. After releasing it online in early 2017, the project gradually grew. Now, 8 years later, the database has processed 100 million drawings. Technically, the backend uses .NET and WebSockets for real-time synchronization, the frontend is handwritten HTML/JS, data is stored in PostgreSQL and MongoDB, and Wasabi cloud storage is used to save costs. The author uses an AI content classification model trained in 2021 to filter inappropriate content. On busy days, there are 30,000 active users with approximately 3.16TB of storage. The 100 millionth drawing was a red balloon. This story shows the transformation from a failed project to a successful product, highlighting the indie developer’s efforts in scaling a real-time multiplayer game, handling reconnection edge cases, and content moderation. The author is open to discussing technical details, such as how .NET handles the load and the application of AI in content safety.
Original Link:Hacker News






AI周刊:大模型、智能体与产业动态追踪
程序员数学扫盲课
冲浪推荐:AI工具与技术精选导航
Claude Code 全体系指南:AI 编程智能体实战
最新评论
朝鲜的互联网基础设施一直是黑箱,这次调查很有价值。光纤网络的物理布局确实能反映很多信息,比如重点区域和网络拓扑。
RSS+AI的组合确实很有价值,信息过载时代确实需要智能筛选。建议增加一下跨来源的内容去重功能,避免重复推送同一话题。
侧边栏调用Gemini的思路很实用,不需要切换标签页就能使用AI。不过想了解一下是否支持自定义API Key,使用官方API可能会有限制。
登录重定向问题确实很烦人,特别是对于刚上线的项目。这个解决方案的思路很清晰,不过不同框架的实现可能需要调整。
注册流程的漏洞分析很有价值,小号入侵是很多平台都面临的问题。建议补充一下防御机制的实现细节,比如设备指纹和行为分析。
ClaudeCode在复杂项目上的表现确实不错,特别是对上下文的理解能力。不过想了解一下生成代码的可维护性如何,是否需要大量人工调整?
小团队确实需要更精简的技术栈,AI优先的思路很有前瞻性。不过团队成员的技术栈可能会比较分散,维护成本如何控制?
云服务的credits使用策略确实容易踩坑,特别是对于第三方模型的限制。建议用户在使用前仔细阅读服务条款,避免浪费额度。