Notes from the build.

How we plan, build and hand over AI products and business systems for companies in Singapore, Malaysia and Indonesia.

AI17 Oct 202618 min read

What Claude does inside our ERP, CRM, WMS, EAM and HR

Claude reads documents, answers from your own files and figures, and drafts messages in all five systems. Checks written in code route every draft, a person confirms it before anything posts or reaches a customer, and every workflow still runs with AI switched off.

Read article →

API

Engineering11 Oct 202618 min

App running costs in Singapore and Malaysia: the third-party bills

Most of what an app costs to run each month in Singapore or Malaysia is other companies’ bills: hosting, maps, SMS, email, payments and AI. In the worked month below, one build decision, SMS login codes to Malaysian numbers, is 90.6% of a US$1,122.50 bill.

GEO

Engineering11 Oct 202616 min

What is generative engine optimisation (GEO)? A guide and checklist

Generative engine optimisation (GEO) is the work of making your pages easy for AI assistants and AI search features to find, understand and cite. It extends SEO rather than replacing it, and most of the work is crawler access and plain writing.

Systems11 Oct 202615 min

WhatsApp Business API pricing in Malaysia and Singapore (2026)

Meta’s rate cards effective 1 October 2026 charge RM0.3467 for a marketing message to a Malaysian number and S$0.0937 to a Singapore number, and RM0.0564 or S$0.0205 for a utility, authentication or service message. Service replies are now free only up to 1,000 a month per number, and in a worked month on one number, a single promotion is 73% of Meta’s fees.

Q&A

AI15 Apr 202518 min

Building document Q&A that cites its sources takes four things

A document assistant earns trust when every answer cites a page the reader can open and it says so when the answer isn’t there. Most of the build goes into preparing the documents, filtering by permissions before the search and testing against real questions.

OCR

AI11 Feb 202516 min

OCR vs AI document extraction: invoices, receipts and expense claims

OCR turns a scan into text, and AI extraction turns that text into the named fields your system can post. To automate invoices, receipts or expense claims you need both, plus checks that catch what a confidence score misses.

SG·MY·ID

Systems10 Dec 20247 min

How we built our own ERP, CRM, WMS, EAM and HR (2023–2024)

We started building the ERP, CRM, WMS, EAM and HR in June 2023, and the fifth went live at the end of November 2024. The CRM, WMS, EAM and HR connect to the ERP, and each country’s tax, leave and inspection rules are dated settings. All five run the same code in the cloud or on-premise and launched without AI.

ERP

Systems15 Oct 202418 min

ERP customisation: what to configure, customise or keep standard

Most ERP changes belong in configuration, because settings carry over at each vendor update and custom code has to be retested. Extend the system only for processes customers choose you for, and keep finance, inventory and tax logic standard.

8→4

Process16 Jul 202410 min

How we embraced AI: building software with Claude

Since June 2024, every custom project that starts with us is built with Claude and runs as a 4-week sprint in place of the 8-week build. Claude drafts code, tests and documentation. The author of each pull request and then the technical lead review every change, and the CI tests must pass before it merges. Data access and scope decisions take as long as before.

FIX

Engineering11 Oct 202214 min

How to rescue a stalled software project in the first two weeks

Rescue a stalled project in a fixed order: get it building from a clean machine and reclaim the accounts in days 1–3, map what exists in days 4–10, and hold the work to a written plan you can check on staging from day 14.

W1–W8

Process9 Aug 202210 min

How we run an 8-week custom software build, week by week

Every custom build we take on runs 8 weeks: requirements in week 1, a clickable prototype in week 2, its revision with users in week 3, the build on real data in weeks 4–7 and deployment with handover in week 8. Three of those weeks come before any build, because a comment on a prototype changes a drawing and the same comment in week 6 changes the database.

MLP

Product14 Jun 202215 min

MVP vs MLP: building first versions people keep using

A minimum viable product tests whether an idea is worth pursuing. A minimum lovable product does one job completely, so the staff or customers who depend on it keep using it, and that’s the version to build once anyone does.

Subscribe via RSS →