Blog 文章
Writing and selected answers. 文章与代表回答。
A compact reading path organized by the areas I have worked in: WeShop, Memex, recommendation systems, and technical foundations for AI applications.
按我的工作范围整理，每组只保留 2-3 篇代表内容：WeShop、Memex、推荐系统，以及 AI 应用相关的纯技术理解。
Reading path 阅读路径
Organized by work area. 按工作范围整理。
Each area keeps only a few representative pieces. The selected essays now keep dedicated English and Chinese editions on this site, with original-source links preserved on the essay pages.
每组只放少数代表作。精选文章会在站内维护中英文两个版本，原始链接保留在文章页里。
WeShop AI and commercial creative workflows WeShop AI 与商业创意工作流
Writing around WeShop's path from digital-model generation experiments to a commercial AI photography and creative-production product.
围绕 WeShop 从电商数字模特技术实践，走向 AI 商拍、商业创意生产和全球化 SaaS 的一组文章与回答。
-
Can AI replace e-commerce models now? AI 技术代替电商模特，现在可以实现了吗？
A widely read WeShop answer on AI models, virtual try-on, e-commerce photography constraints, cross-border merchandising, and why usefulness can arrive before perfection.
一篇关于 WeShop 场景的高赞回答，解释 AI 模特、虚拟试衣、电商商拍约束、出海商品图和生成式 AI 的可用性边界。
Read essay 阅读全文 -
What Building WeShop Taught Me About Generative-AI Products 谈谈做WeShop过程中对AIGC产品的一些思考
A core WeShop product essay on defining an AI product before the technology is mature, plus compute cost, model evaluation, and why real customers matter early.
WeShop 早期最重要的产品思考之一，讨论用户反馈、PMF、质量问题和 AIGC 产品为什么要尽早接触真实客户。
Read essay 阅读全文 -
We Released WeShop Commercial Photography 1.5 — Some Recent Reflections 我们发布了WeShop商拍1.5版----分享一些最近的思考
A product-iteration essay on preserving product details, changing backgrounds, quality limits, and how AI commercial photography expands from demos into workflows.
WeShop 1.5 的产品迭代复盘，讨论商品细节、背景替换、质量边界，以及 AI 商拍如何从 demo 进入真实工作流。
Read essay 阅读全文
Memex, agents, and personal AI Memex、Agent 与个人 AI
Writing around local-first personal memory, AI agents, open-source growth, and using AI coding to build a real product.
围绕 Memex、local-first 个人记忆、AI Agent、开源增长和 AI coding 产品实践的一组内容。
-
We Built Something About Recording Yourself, Then Decided to Open Source It 我们做了一个关于“记录自己”的东西，然后决定把它开源
The product thesis behind Memex: personal records, private context, memory, and why local-first trust matters in the AI era.
Memex 的产品原点：个人记录、私人上下文、长期记忆，以及 AI 时代为什么重新需要 local-first 信任。
Read essay 阅读全文 -
Why Record Yourself So Often? 为什么一定要频繁记录自己？
A public-facing explanation of personal records, AI anxiety, memory, agency, and why private context becomes useful again in the agent era.
把 Memex 的产品理念讲给更广泛读者：个人记录、记忆、隐私、AI 焦虑和私人上下文。
Read essay 阅读全文 -
How an Open-Source Project With No Resources Got Discovered 一个没资源的开源项目，是怎么被人看见的
A postmortem on Memex's open-source cold start: useful work, public discussion, timing, and distribution without a large existing audience.
Memex 开源冷启动复盘：没有预算、没有合作，如何靠真实作品、有效讨论和时机被看见。
Read essay 阅读全文
Search, recommendation, and earlier machine learning 搜索、推荐与早期机器学习
Earlier writing from search, recommendation, advertising, and industrial machine-learning practice before the current generative-AI wave.
这一组代表更早期的搜索、推荐、广告和工业机器学习实践，是进入生成式 AI 之前的重要经验底色。
-
A ranking-systems postmortem: algorithm engineers are engineers first 论算法工程师首先是个工程师之深度学习在排序应用踩坑总结
A 1,000-plus-like essay from the Mogujie search/recommendation period, arguing from real ranking-system work that algorithm engineers need strong engineering judgment.
蘑菇街搜索推荐阶段的一篇过千赞专栏，用深度学习排序项目的真实踩坑说明：算法工程师首先得是工程师。
Read essay 阅读全文 -
Reading Airbnb's KDD 2018 best paper from a recommendation-systems perspective 如何评价Airbnb的Real-time Personalization获得2018 kdd最佳论文？
A high-engagement answer on Airbnb's embedding-based search ranking paper, mixed with practical reflections from domestic recommendation-system work.
关于 Airbnb embedding 搜索排序论文的高赞回答，也夹带了不少国内推荐系统实践的经验判断。
Read essay 阅读全文 -
What is new in reinforcement learning for recommender systems? 增强学习在推荐系统有什么最新进展？
A high-interaction answer using YouTube's RL-for-recommendation papers to explain long-term reward, exploration, large action spaces, and engineering constraints.
用 YouTube 推荐系统里的强化学习论文，解释长期收益、探索、大规模 action space 和推荐场景里的工程约束。
Read essay 阅读全文
Generative AI and application foundations 生成式 AI 与应用基础
Technical and product writing from the early generative-AI transition, before and during the move into WeShop.
这一组是生成式 AI 早期转向时的技术和应用理解，也解释了为什么后来能进入 WeShop 这样的产品实践。
-
A Practical Guide to Diffusion Models Diffusion Models 导读
A technical primer for readers entering image generation, useful as a marker of the technical base behind later commercial AI image work.
面向图像生成入门读者的 Diffusion Models 技术导读，也是理解后续 AI 商拍实践的技术背景。
Read essay 阅读全文 -
ChatGPT's Core Technologies, Seen From the Application Layer 应用视角下 ChatGPT 背后的关键技术讨论
A technical-product discussion of ChatGPT from the perspective of application builders, not only model observers.
从应用构建者视角讨论 ChatGPT 背后的关键技术，而不是只停留在模型观察。
Read essay 阅读全文 -
If the Agent Loop Is Tiny, What Are the Other Tens of Thousands of Lines For? 在研究编程 Agent，Agent 核心就几十行代码，那剩下的几万行到底在解决什么问题？
A production-engineering answer on why real agents need tools, permissions, context, validation, UI, logs, recovery, and product structure.
从工程角度解释生产级 Agent 为什么不只是几十行循环，还需要工具、权限、上下文、校验、UI、日志和恢复机制。
Read essay 阅读全文