Every deck is rewritten. A pre-course interview turns the cases, the exercises and even the vocabulary into situations your people actually meet — not a generic course with a new cover page. Taught by someone who maintains five production AI pipelines, so "what actually happens in practice" gets a real answer — including where it fails.
Management said everyone should take an AI course. Everyone did. Back at their desks, the work runs exactly as it did before.
The trainer taught generic tool operation. The moment somebody asked "but how would that work with the way we quote?", the answer stopped.
The slides are sitting in an inbox and nobody has opened them again. Three months on, actual AI usage in the company is still zero.
Three companies in three unrelated businesses. Same course, three different versions of it.
The classroom version of the same method ran in the education section of Mandarin Daily News — an electroplating plant, an equipment maker and a primary-school classroom, all the same job. This is not a course for one industry. If people in your company spend the day typing, looking things up, writing quotes and answering customers, there is something here worth teaching. Your industry not on the list? That is what the pre-course interview is for — the examples get swapped for yours.
"AI applications" means something different to a metal fabricator than to a publisher — the cases, the exercises and the vocabulary should all differ. That is the whole method here.
The question is not "which topics would you like". It is which part of the work takes longest, goes wrong most often, and nobody wants to do. That is what the session solves — not a tour of tools.
Usually 30 to 60 minutes, with the people who actually do the work, not only their managers.
Cases become your business scenarios, exercises use your real documents, and the sample prompts are written in your trade's language. The custom decks delivered so far run to roughly fifty pages each — not a shared slide library.
If you would rather not share internal material, equivalent cases are built from public sources.
A session can be built so people leave with a small working tool made for your industry — a quotation calculator, a price-trend estimate. That is how it has been delivered before, and the tool stays with the company.
Anyone can hand out slides. Only something that runs keeps getting used.
None of this is read second-hand. It has been built, shipped and broken in production — which is why the failure modes get airtime, not just the success stories. Topics can be picked, combined or taken deeper.
The most requested session. Start from your actual processes, map which steps can save time today and which are a waste of effort, and work through them in the room.
Suits whole-company or single-department training. No technical background needed.
Plain-language explanation of why it gets things wrong, why it invents things, and why the same question gives two different answers. The goal is not technical literacy — it is immunity to sales talk.
Particularly useful for decision makers: whoever signs off the budget should know what they are buying.
How to make an assistant answer only for your company and stop it inventing. How to write the knowledge base, how to fix a wrong answer, and when it should hand over to a human. This is RAG in practice, explained without touching code.
Customers have started asking ChatGPT who is good in an industry. Whether your name appears in that answer takes a different kind of work from traditional SEO.
Plenty of people talk about this. Most of it is guesswork. Here there is measured data.
What happens when staff paste a customer list into a chatbot, whether a public-facing assistant can be talked into saying things it shouldn't, and why a "read-only" setting may not be locking anything at all.
The topic companies worry about most, and the one with the fewest people able to teach it from experience — because to have the examples, you have to have been hit.
AI video, AI copywriting, automated lead-finding — what they can genuinely do today. The session deliberately concentrates on where they fail, because that is where budgets get wasted.
Aimed at marketing teams and at managers who have to decide whether to adopt anything at all.
The version for teachers. Not theory — working demonstrations: turning an existing textbook passage into a group performance, a puzzle trail or a quiz round; using AI to inventory a neighbourhood and thread the chosen stops into a themed walking lesson; letting pupils get scammed once inside a safe simulation and then trace which step they lost it at.
The core point is identical to the corporate sessions: what AI produces is a draft, and the judgement stays with the teacher. It will pitch the tone too frightening, set questions too hard or too easy, put a pin in the wrong place on the map — so a person always closes the loop.
Still quoting from a veteran's instinct? Only noticing raw-material price swings after the fact? This session covers turning the history you already hold into an estimate that can be read — and justified.
Fits manufacturing, wholesale and trading, and any business whose quotes track a market.
Not a motivational story — a working method. One person maintaining dozens of live projects: which steps AI genuinely took over, and which were tried and handed back to a human.
For software teams, startups, and owners weighing up whether headcount really has to grow.
Talks for trade associations, chambers of commerce, conferences and company-wide events. Including less conventional subjects — simulating a city's public opinion with a synthetic population model, for instance. Not suitable as a regular course; excellent as a talk.
Taking the most requested format — three hours — the session runs in seven parts. The first four are common ground. The fifth comes out of the pre-course interview, differs for every company, and is where the value of the whole thing sits.
01 · Meeting AI again, properly
What it is actually doing, how input tools have evolved through four generations, what it is and isn't good at. Demonstrated live in the room rather than described.
02 · How to talk to it
Treat it as a bright assistant on their first day. The four elements of a good instruction — context, task, format, example; why a conversation is iterative rather than one-shot; and how to get the model to draw the instruction out of you when you can't phrase it. Image generation is covered here too.
03 · It gets things wrong, fluently
Models state errors with complete confidence. The expensive mistake isn't the error — it's the error that reads well and nobody checks. Then the three data rules: not the client's confidential material, not the company's crown jewels, de-identify before use.
04 · Everyday use cases
The six that come up most in an office: business correspondence, meeting notes, quotations and internal approvals, translation and editing, spreadsheet work, SOPs and training material. Each demonstrated, not listed.
05 · Your company's own scenarios — customised; different every time
The product of the pre-course interview. Your real workflow and real documents on the table, worked through together. There is no off-the-shelf version of this part — it is written for you.
06 · Where this goes next, and what it means for you
How far AI has come in your particular industry, which directions it will reach your day-to-day from, and why the gains will not be evenly distributed. Ends with an adoption roadmap you can actually follow.
07 · Discussion
Four questions that help each department find its own starting point. Not a Q&A to close on — a working session to get the answers said out loud.
Appendix · Six AI assistants compared
Claude, ChatGPT, Gemini, Grok, Perplexity and Meta AI — what each is good at, how they differ, and which to start with. No vendor is being sold here, only how to choose.
The real difference between this and other AI courses isn't the material — it's who is standing at the front.
I'm not a trainer assembling second-hand material. I maintain AI systems that people genuinely use every day:
they break, they get asked things they can't answer, they cost money.
So when you ask what actually happens in practice, the answer comes from someone who has done it, not from
a citation. For the same reason I'll say plainly which things can't be done yet and which investments will be
wasted — the sort of sentence a trainer who lives off course fees cannot afford to say.
A 30-minute call first, to pin down what you actually need solved — then the format and the hours follow from that.
Where you're stuck, who should attend, what the budget and timing look like. By the end of that call it's clear whether this is worth doing — and if it isn't, you'll be told.
Interviews with the people doing the work, then the deck is rewritten. You see and approve the outline along the way — nothing is a surprise on the day.
On-site or online. Hands-on rather than one-way; questions raised in the room are written down as they come.
Material, the questions from the day, and a suggested next step. If you decide to go further, adoption is a separate conversation — and if you don't, the session stands on its own.
All three corporate clients agreed to be named. Here is what was actually delivered.
Raw-material prices moved constantly, quoting depended on the instincts of long-serving staff, newcomers couldn't pick it up, and explaining a number to a customer was hard.
WHAT WAS DONE A full deck rewritten for the industry, plus two working tools — a price-trend estimate and a quotation calculator — used as teaching aids and left with the company afterwards.
The company wanted to adopt AI but didn't know where to begin, and worried about buying tools nobody would use.
WHAT WAS DONE A full in-house course built around that company's actual workflow, mapping in the room which steps were worth doing and which to leave alone for now.
The people approving budgets usually know AI through headlines and vendor decks, which is no basis for judging which promises are real.
WHAT WAS DONE No code. The mechanism explained until it lands, weighted towards when it fails and which claims should trigger suspicion.
Articles in the education section of Mandarin Daily News on practical classroom uses of generative AI. Founded 1948; its readers are primary-school teachers and parents. Articles are in Chinese.
Repackaging classroom material with prompts 2026-07-29
One textbook passage, three formats: a script for group performance, a quiz round, and a clickable web game. The material doesn't get rebuilt — the lesson just changes shape.
Planning a neighbourhood walking lesson with AI as a co-planner 2026-08-05
A teacher doesn't have to be a local historian: have AI inventory the area, thread the chosen stops into a themed route with an observation task at each, then shape what the pupils bring back into a discussion. AI spreads out the options; the teacher chooses.
Building a scam-awareness drill pupils can fail safely 2026-08-12
Online-safety talks happen every year and pupils still click the link. Three layers on one piece of material: read a story and find the step where it was lost, then argue live with an AI playing the scammer, then a five-round scored web game where a wrong answer shows the consequence. The finished game is playable.
From $80 per hour (NT$2,500). The final number depends on teaching hours, group size and how far the material is customised.
A 30-minute call comes first to establish what you need solved. That call is free.
Mandarin by default — most sessions are for teams in Taiwan. Written material can be produced bilingually.
If you need the session itself delivered in English, raise it before booking. It changes preparation and scheduling, so it's worth a conversation first rather than an assumption.
Yes. On-site within Taipei City carries no travel charge; elsewhere in Taiwan travel is billed separately.
Online works too, but if the session includes hands-on exercises, in person is noticeably better.
Not at first — that's what the interview is for. Every deck is rewritten afterwards, so the cases and exercises are situations your people actually meet.
Full custom material has been produced for three different industries: metal finishing, air-pollution-control equipment and precision machining. A different industry isn't a problem as long as the workflow can be explained.
Yes, and they are the main audience. There is no coding.
The point is a correct mental model: what AI can and can't do, when it goes wrong, and which claims are sales talk. None of that needs a technical background — it needs someone to explain it clearly.
You can, and it's worth doing — real data changes how the room responds.
Scope and confidentiality are agreed during the interview, including how material and data are kept or destroyed afterwards. Not sharing internal data is completely fine; equivalent cases get built from public sources.
Yes, recording rights are agreed as part of the booking.
One honest caveat first: recordings are watched far less often than people expect. Rather than pay for a video nobody opens, put the budget into the written follow-up — that does get referred back to.
No minimum. Sessions have run for five people and for fifty.
That said, for a very small group the AI consulting service is usually better value — two hours one-to-one plus a written action plan. You'll be told which one fits when we talk.