https://du-xiang.com machine learning, production systems, and other small obsessions. full-text essays, no tracking, dotted borders. en-us Movable Type 3.2 (allegedly) https://www.rssboard.org/rss-specification https://du-xiang.com/blog/on-feeling-weak https://du-xiang.com/blog/on-feeling-weak Wed, 10 Jun 2026 01:20:00 -0700 Twenty years of learning, four years of acceleration, and the question of what remains ours. A personal essay, written late at night, in both my languages. https://du-xiang.com/blog/agent-teams-architecture https://du-xiang.com/blog/agent-teams-architecture Mon, 09 Feb 2026 09:00:00 -0800 AI agents that talk to each other, peer-to-peer. How that compares to hub-and-spoke sub-agents, and when each one earns its complexity. https://du-xiang.com/blog/skills-vs-mcp-agents https://du-xiang.com/blog/skills-vs-mcp-agents Wed, 04 Feb 2026 09:00:00 -0800 How should a machine hold its tools? Enterprise cases, security considerations, and some opinions. https://du-xiang.com/blog/context-engineering-memory https://du-xiang.com/blog/context-engineering-memory Thu, 07 Aug 2025 09:00:00 -0700 Giving language models a memory, so they stop forgetting you between conversations. From short-term buffers to knowledge graphs. https://du-xiang.com/blog/seo-geo-content-generation https://du-xiang.com/blog/seo-geo-content-generation Fri, 20 Sep 2024 09:00:00 -0700 Teaching machines to write things other machines want to read. The circle of life, automated. https://du-xiang.com/blog/realtime-quick-suggest https://du-xiang.com/blog/realtime-quick-suggest Mon, 10 Jun 2024 09:00:00 -0700 Autocomplete fast enough to finish your sentence politely. FAISS, WebSockets, FastAPI. https://du-xiang.com/blog/multi-lora-serving https://du-xiang.com/blog/multi-lora-serving Fri, 15 Mar 2024 09:00:00 -0700 Sixteen fine-tuned models sharing one graphics card, like graduate students sharing an apartment. vLLM, PEFT, SageMaker. https://du-xiang.com/blog/latent-reasoning-slm https://du-xiang.com/blog/latent-reasoning-slm Thu, 28 Dec 2023 09:00:00 -0800 Chain-of-thought quality without the token toll. Small models, larger ambitions. https://mp.weixin.qq.com/s/RryxBWAClB9AgIqFvoSbhg https://mp.weixin.qq.com/s/RryxBWAClB9AgIqFvoSbhg Thu, 01 Jul 2021 09:00:00 -0700 Data science applied to Taobao lipstick trends. Off-site, and in Chinese. https://mp.weixin.qq.com/s/Od3sdJ60wn4WKgM2PfC05A https://mp.weixin.qq.com/s/Od3sdJ60wn4WKgM2PfC05A Sun, 03 May 2020 09:00:00 -0700 Web-crawled earthquake data with heatmap visualization. Off-site, and in Chinese.