Source:
ChatGPT:
1. What the Straits Times article says
The Straits Times article was published on 26 September 2026 and profiles Hosan Swee, a Singaporean engineer and AI educator who built KiasuMiles despite not having traditional software-development skills. (The Straits Times)
His problem was familiar to serious miles chasers: he and his wife had around a dozen cards, each with different:
bonus categories
minimum-spend requirements
monthly caps
exclusions
payment-method rules
merchant/MCC treatment
Remembering all of this at the point of payment became cumbersome.
His solution was to build an AI-powered recommendation system that answers questions such as:
Which card should I use at Cold Storage with Apple Pay?
or
Which card should I use for GrabFood?
The important difference from simply asking ChatGPT
KiasuMiles isn't intended to be a generic chatbot that guesses an answer.
It has a structured database of Singapore credit-card rules, currently covering roughly:
50 credit cards
3,300+ merchants
The system matches the user's cards against the merchant, spending category and payment method, while taking into account caps and conditions. (The Straits Times)
For example, the article uses Cold Storage and says KiasuMiles may identify the DBS yuu card, while also warning about its conditions such as the monthly spending requirement and participating yuu merchants. (The Straits Times)
2. The really interesting part: AI did the coding
This is arguably the bigger story than the miles tool itself.
Swee had the idea but wasn't a conventional programmer. He used:
ChatGPT
Codex
ChatGPT Work
to turn natural-language instructions into working software.
One of his prompts essentially told the AI to build a structured knowledge base containing:
card → earn rate → minimum spend → categories → caps → exclusions.
The AI then gathered information from bank and merchant websites and structured it.
He says something that is quite important for the current AI era: the AI is not replacing the need to think about the product; it is dramatically reducing the cost of implementation. (The Straits Times)
He also built it incrementally rather than asking AI to create everything in one giant prompt, testing the recommendation logic with realistic scenarios before expanding it.
3. Why KiasuMiles is potentially more useful than asking ChatGPT directly
This is the strongest technical point in the article.
A normal ChatGPT question might be:
"What's the best card for Cold Storage?"
The problem is that a general-purpose AI may have:
outdated card rules
incomplete information
incorrect MCC assumptions
forgotten spending caps
incorrect exclusions
incomplete payment-method rules
KiasuMiles instead maintains a specialised knowledge base and uses that to produce the recommendation. (The Straits Times)
The current KiasuMiles site also says it returns the card from your own wallet, rather than simply recommending the theoretically best card available in Singapore. (KiasuMiles)
That's an important distinction.
4. What KiasuMiles actually looks like now
The product has evolved beyond the description in the newspaper article.
Its current site describes a workflow where you connect it to a compatible AI agent and tell it which cards you own. It then answers:
Merchant + payment method → your eligible cards → best card + conditions + fallback. (KiasuMiles)
Its open-source GitHub project describes the same concept as a Singapore credit-card rewards optimisation MCP. (GitHub)
The project's published architecture is interesting because the hosted service says it doesn't require card numbers, expiry dates, CVVs or banking credentials. It works with card names and reward rules rather than payment credentials. (GitHub)
The current project reports 48 cards and about 3,308 merchants, which is broadly consistent with the figures reported by the article. (GitHub)
5. What the online miles community is saying
I searched specifically for discussion around KiasuMiles + the article, rather than treating general Singapore credit-card discussions as reactions to this particular story.
Important caveat
The article is only a couple of days old, so there isn't yet a huge amount of independently indexed discussion.
I could find much more substantive material around KiasuMiles itself than around the article on Facebook/Instagram/X/Threads/TikTok.
That means I wouldn't claim there is already a broad social-media consensus.
The reaction so far is better understood through several themes.
π’ Theme 1: “This is exactly the problem miles hackers have”
This is the most obvious appeal.
Singapore's credit-card ecosystem has become sufficiently complicated that even experienced miles collectors can make mistakes.
The creator himself explains that the losses aren't necessarily one giant mistake; they're lots of small incorrect taps over time. His own earlier write-up describes discovering that he was sometimes using the wrong card despite following the miles community. (Hosan's Substack)
The current KiasuMiles product page makes the same argument visually:
many cards → one cashier → a few seconds to decide. (TinkerHaus)
That is a very Singapore-miles-community problem.
π’ Theme 2: The interesting innovation isn't “AI recommends a credit card”
This is where I think the article has more significance than it initially appears.
People have been asking ChatGPT questions like:
"Which card gives the most miles for dining?"
for a while.
The more interesting idea is:
AI + structured data + personal wallet + merchant database
rather than simply:
AI chatbot + question
The current KiasuMiles architecture even breaks the recommendation into steps such as:
identify merchant → determine MCC → filter user's wallet → rank eligible cards → return recommendation. (TinkerHaus)
That is much closer to an actual decision-support system than a chatbot.
π‘ Theme 3: Accuracy is going to be the biggest concern
This is the biggest weakness that experienced miles users are likely to focus on.
Credit-card rewards change constantly.
Banks can change:
earn rates
minimum spend
caps
exclusions
MCC eligibility
wallet/payment-method treatment
merchant partnerships
So the question isn't merely:
"Can AI build the database?"
It's:
"Can the database stay correct?"
KiasuMiles says it regularly updates its rules, and the creator explicitly warns that simply asking a general AI chatbot can produce incomplete or outdated information. (The Straits Times)
That makes data maintenance and verification arguably more important than the AI itself.
π‘ Theme 4: MCC/payment-method edge cases remain difficult
This is an especially important issue for Singapore miles collectors.
For example:
Merchant → payment method → payment processor → MCC → card rules
can all affect the actual earn rate.
Apple Pay isn't necessarily equivalent to every other payment method.
Online transactions aren't necessarily equivalent to physical transactions.
A merchant's actual MCC may differ from what consumers expect.
The KiasuMiles approach attempts to explicitly incorporate merchant category and payment method rather than simply saying "use Card X for groceries." (The Straits Times)
That is probably one of the strongest aspects of the idea.
π’ Theme 5: Open-source makes it more interesting
The project is publicly available on GitHub under an MIT licence. (GitHub)
That potentially allows developers/miles enthusiasts to:
inspect the implementation
identify bugs
contribute improvements
see how the recommendation engine works
build their own integrations
The creator's Tinkerhaus page describes it as an open-source AI card-recommendation skill. (TinkerHaus)
That gives it more credibility than a mysterious black-box "AI miles optimiser."
6. Facebook / Instagram / X / Threads / TikTok
I specifically searched these platforms for the article and KiasuMiles.
The indexed public discussion is currently quite thin, especially compared with Reddit and specialist Singapore miles communities.
That's important because social-media search engines often don't expose all posts, especially:
private Facebook groups
Instagram comments
Threads posts
TikTok comments
X posts behind indexing/login restrictions
So I wouldn't interpret the lack of indexed posts as "nobody is interested."
What is visible points to the story being picked up primarily as an AI/product-building story, rather than becoming a major controversy around credit-card rewards.
7. The bigger discussion this article has triggered
I think there are actually two stories hidden inside this article.
Story A — Miles
Singapore's credit-card optimisation game is becoming so complicated that there is now a need for a dedicated AI decision engine.
That's a sign of how sophisticated the local miles ecosystem has become.
Story B — AI
A non-traditional programmer can now go from:
"I wish there was a tool that did this"
to
a functioning specialised financial application
using AI coding tools.
That second story is arguably more significant.
The creator himself says that learning basic software-development terminology improved his ability to communicate with AI. In other words, you don't necessarily need to become a professional programmer; you need enough technical understanding to direct, test and challenge the AI. (The Straits Times)
8. My overall read of the article
I'd summarise it this way:
The article isn't really about getting more miles. It's about the transition from "AI answers questions" to "AI operates specialised decision systems."
KiasuMiles takes a particularly Singaporean problem — "Which of my 8–12 credit cards should I tap right now?" — and turns it into a structured AI workflow.
The strongest parts are:
Personalised: only considers cards you actually own.
Merchant-specific: rather than generic category recommendations.
Condition-aware: considers caps/minimum spends.
Payment-aware: attempts to account for payment method.
Structured: relies on a maintained database rather than pure chatbot memory.
Open-source: implementation can be inspected. (KiasuMiles)
The biggest thing I'd watch is data accuracy. A miles tool can be 95% right and still give you a wrong recommendation at exactly the merchant where the remaining 5% matters. Credit-card T&Cs also change frequently.
And there's a funny Singapore angle
The name KiasuMiles is actually quite appropriate.
The traditional miles enthusiast workflow is:
carry 8 cards → remember 20+ rules → check Telegram/Reddit/blogs → remember monthly caps → remember MCC → stand at cashier → panic → tap card → later check whether you got the right miles.
KiasuMiles tries to turn that into:
"I'm at Sheng Siong. Which of my cards?" → one answer.
Its own current demo illustrates this with merchants such as Sheng Siong and Shell, showing the recommended card, earn rate, cap and fallback. (TinkerHaus)
Given that you already use a fairly sophisticated Singapore miles stack involving Citi Rewards, UOB cards, Amaze/HeyMax etc., this is actually much closer to your real-world use case than a generic credit-card comparison site.
No comments:
Post a Comment