# Case Files

The problem

No IT department at a brand-new company — how do you have systems that can handle orders from day one?

What a new company lacks most is rarely orders — it's a system that can handle them. No IT department, no budget for implementation consultants, and operations run by me alone. So I built it myself, with AI.

How I did it with AI

1. ①

   Product data pipeline

   Cleaned and standardised a supplier catalogue of about 1,500 SKUs, and turned the "cost → exchange rate → tax → margin" pricing matrix into a re-runnable script. Change one parameter and the whole price list recalculates.

2. ②

   Self-built ERP

   Deployed Odoo myself for quotes, orders and inventory, linking "supplier → warehouse → customer" into one continuous order flow.

3. ③

   B2B product store

   Built on Odoo and organised by application (workholding, cutting tools, measuring, machine protection, handling), with HS codes and landed-cost notes on every product page.

The problem

Going global isn't hard because of selling — it's the hundred pieces of paper after the sale.

Importing into Malaysia, I dealt with these three kinds of work every day. The usual answer is to hire or outsource. I turned them into tools.

How I did it with AI

1. ①

   SDS generation

   Enter the product data and a compliant safety data sheet is generated. It used to mean searching databases and formatting by hand — now it means filling in a form.

2. ②

   Import tax checks

   Used AI to read Malaysia's official gazettes directly — the customs duties order, sales tax orders and exemption orders, 300+ pages in all — checking duty, sales tax and exemptions HS code by HS code. It also untangled regimes that are often confused: low-value goods tax vs. sales tax on commercial imports, where De Minimis actually applies, and why exemptions must be approved before the goods arrive.

3. ③

   Customs & trade documents

   English invoices, HS codes, country of origin and Incoterms (who is liable under DAP vs. DDP) — all templated and filled in automatically.

The problem

Tens of thousands of years ago, people painted survival lessons on cave walls for the next generation. What can a dad use today?

There was no writing yet — only bison, spears and hunting formations on the wall. Not for beauty, but so the next generation would know how to survive. EduStory is the same thing.

Every episode starts from two real triggers: I see my kid Joey do something, and I know where it is likely to lead — because I've made the same mistake myself and paid for it. A child's real behaviour, a father's real observation, a father's real lesson: that is all the story is made of.

How I did it with AI

1. ①

   Don't lecture — turn the habit into a character

   A marker left uncapped sneezes ink everywhere; toys thrown on the floor run away at midnight. What the child sees isn't blame but consequences — and then decides for himself whether to change.

2. ②

   An animation crew made of AI

   I split the producer, scriptwriter and artist into three AI agents, each with its own spec, quality-gate rules and a shared character bank. Each episode is 3 minutes and 18 shots, and the family looks the same in every one.

3. ③

   Iterated like a production line

   Episode one was 1 minute and 8 shots — it felt like a slideshow. After reviewing it second by second, I set the rules: one story per episode, no more than 25 characters of dialogue per shot, five seconds left for action and breath. By episode four the pacing finally settled.