About
I am a senior AI researcher at the Agency for Science, Technology and Research (A*STAR), Singapore. I work on applied LLMs and agent systems — in the trenches, not the press releases — for problems where the agent has to actually ship and stay shipped.
Before A*STAR I spent seven years at Accenture Singapore as a senior consultant, working on enterprise systems across MedTech, FinTech, and supply chain. Before that, twenty years of IT in roles that taught me what production actually looks like, which is mostly different from what the slide decks claimed it would.
What I write about
Agent architecture. The thesis I keep returning to is simple enough to fit on a back cover: the model spends; the harness budgets. The longer version: agent architecture is the design of bounded resource allocation under uncertainty. Twenty-seven patterns, seven cognitive functions, six topologies — all of them are different ways of allocating a finite resource pool under different scarcity constraints.
My forthcoming Manning book, Designing AI Agents (MEAP May 2026), is the long form of that argument. The newsletter, Agent Design Patterns, is where the shorter pieces live.
The other half
I am also a certified counsellor — trained under China’s national psychological counsellor certification — with seven years on a Singapore counselling helpline. Five hundred clinical hours, all phone work.
Phone counselling is reading a person through a narrow channel: voice, pause, breath, the words they pick and the ones they avoid. You decide call by call when to escalate and when to hold. That skill set turns out to map onto the harness around an LLM more directly than any of my software engineering background did. A counsellor and a harness do the same job — not produce the right answer, but recognise the wrong one before it commits.
Books
I have written six Chinese-language books on machine learning, GPT, AI agents, RAG, and data analysis — the cumulative readership is in the hundreds of thousands, and the ML book has held a 9.0 rating on Douban for several years. Two English books are forthcoming: Manning’s Designing AI Agents (MEAP May 2026) and Packt’s RAG from First Principles (MEAP mid-May 2026).
Teaching
I teach four columns and video courses on Geek Time (极客时间), China’s leading tech-education platform: machine learning, LangChain, AI applications, and a free ChatGPT public course. Cumulative learners are around 30,000.
Contact
Email is fine but slow. I read everything; replies are selective. The better way to reach me is to comment on a Substack essay or LinkedIn post — public discussion is where I am actually present.
Pen name: 咖哥 (KaGe). The Chinese books, the Geek Time courses, the WeChat following — all under that name. Look me up on the Chinese version of this site if you want the full picture.