PropTech · Singapore

In property advice, a plausible number is a liability.

kaysera builds the systems behind Singapore property advice: the calculators, the planning tools, and the report a client actually takes away. Every rule comes from IRAS, MAS, CPF or HDB, written as tested code and dated to its source.

How we build Start a conversation

§ 01

First client

REDEMPTION is Christian Oh's property practice at JNA Real Estate. Its analysis lived in one spreadsheet, its client decks were screenshots pasted into Canva, and its regulatory assumptions lived in one person's head.

15
public calculators, one shared engine
266
unit tests, all passing
4
regulators' rules encoded and dated

Verified against source · 2026-07-30

The long-term wealth projection calculator on redemption.sg. A form on the left holds purchase price, buyer profile, loan-to-value, interest rate and CPF used. On the right, three figures headed 'Where you stand at year 20' read S$2,339,759 with the property against S$802,027 if the money stayed in savings, above a chart plotting both paths to 2046.

Fig. 1 · Public calculator

One of fifteen calculators, each its own page on the same engine. The inputs sit beside the position they produce, and both dated figures on this screen name their source: the interest rate is prefilled from the published SORA benchmark, re-read nightly, and the stamp duty carries the IRAS rates it was last checked against. It is a public page, so open it and change the numbers.

The report view of the planning tool, showing slide 46 of 53 in a deck titled 'The sale and purchase timeline'. Rows list starting cash plus CPF, the tightest cash point, the date the money is ready, stamp duty, and what is left after every payment. A toolbar above the slide offers Copy deck link, Download PPTX, Print or save PDF, and Present.

Fig. 2 · Client report

The planning tool takes a client's position once and carries it through every relevant calculation. This deck is generated from that single run, not assembled by hand, so no slide can disagree with the calculator behind it, and the same screen exports the file the client takes away. It is internal to the practice, which is why there is nothing to open.

A regulatory engine
Stamp duty, loan servicing ratios, CPF accrued interest, tenure limits and rental yield — every rule pinned to a dated IRAS, MAS, CPF or HDB source, and held there by its own test.
Fifteen public calculators
Each its own page, each running on the same engine, so no two of them can disagree about the same rule.
A planning tool
Takes a client's position once, carries it through every relevant calculation, and generates the written report and the slide deck from that single run.
A nightly rebuild
Re-reads the published benchmark interest rate every night, so no page carries a stale figure into a client conversation.

The spreadsheet-to-screenshot-to-Canva pipeline was replaced end to end. It is live at redemption.sg.

§ 02

Why general-purpose AI stalls here

Singapore property advice is arithmetic wrapped in regulation, which is the shape of work a language model is worst at on its own.

A wrong figure is not a rough draft.
ABSD charged on the wrong profile, a TDSR that clears when it should not, a CPF refund left out of the sale proceeds. The number goes to a client who acts on it. There is no version of it that is close enough to tidy up afterwards.
The rules are public, dated, and they move.
IRAS duty bands, MAS servicing ratios, CPF interest, HDB tenure limits: all published, all revised. A model answering from what it absorbed months ago will quietly serve last year's rules, and nothing in the answer will say so.
You cannot show your work.
When a client asks how a figure was reached, that a model produced it is not an answer. An adviser needs an unbroken line from the number on the page back to the published rule that produced it.

§ 03

Deterministic where it counts. AI where it helps.

The split is the method. Most of the disappointment with AI in this sector comes from putting the work on the wrong side of that line.

Rules become tested code, not prompts.
Anything a regulator publishes — duty bands, servicing ratios, accrued interest, tenure limits — is ordinary code, with a test per rule and a dated citation to the source it came from. It cannot drift. When the rule changes, a test fails.
Every figure carries its provenance.
A rate on a page knows where it came from and when it was last checked. Where a benchmark is published on a schedule, the system re-reads it on a schedule rather than trusting a number typed in once and forgotten.
AI does what AI is good at.
Reading the valuation, the CPF statement or the loan letter a client actually sends. Drafting the narrative around a set of numbers. Turning a scenario into something a person can follow. Language and judgment, not arithmetic that has a correct answer.
The deliverable comes out of the engine.
Reports and decks are generated from the same computation that produced the figures, so the document a client takes away cannot disagree with the system behind it.

The longer version, with worked examples: how we build.

§ 04

Who you would be working with

kaysera is a one-person practice, by design. Whoever works out your rules is whoever writes the code that encodes them, with no handover in between for the detail to go missing. That caps how many engagements run at once, which is the point rather than a limitation to apologise for.

§ 05

Working together

If you advise on Singapore property and the numbers have to hold up, the first conversation is short: what you do today, where the manual work sits, and whether any of it is worth automating. Some of it usually is not.

hello@kaysera.io