FlatFair (by BRickLab)

Inspiration

Buying an HDB resale flat is one of the biggest financial decisions many Singaporeans will make, but answering a few basic questions can still be surprisingly difficult:

Is this flat fairly priced? Can I actually afford it? And where might prices go next?

Singapore has one of the most data-rich public housing markets in the world, yet much of that information is still presented as individual transactions, historical charts, or broad market statistics. For a first-time buyer, translating all of that into an actual decision is not easy.

With resale prices rising and the housing landscape continuing to evolve following the 2024 flat classification changes, we wanted to turn Singapore's open housing data into something more actionable.

That became FlatFair: a data-driven companion for making more informed HDB resale decisions.

What it does

FlatFair helps a homebuyer move from "I'm looking at this flat" to "I understand whether this flat makes sense for me."

Users can:

Explore historical HDB resale price trends across towns and flat types
Enter the characteristics of a flat and estimate its fair market value
Compare it against relevant historical transactions
Understand how affordable the flat is relative to their budget
View a 6-month outlook for prices in the surrounding market

Instead of stopping at descriptive charts, FlatFair connects market analytics directly to an individual homebuying decision.

Our goal is to answer three questions in one place:

What is it worth? Can I afford it? What could happen next?

How we built it

We built FlatFair around Singapore's open housing data, with HDB resale transactions forming the core of our analysis and additional public datasets providing demographic and market context.

The data flows through an end-to-end Databricks pipeline:

Open data → ingestion → Delta tables → cleaning and feature engineering → machine learning → analytics → user-facing application

We use Databricks to process and transform resale transactions, create historical market features, train and evaluate our models, and serve the resulting insights to the application.

For the fair-value model, we use characteristics such as location, flat type, floor area, storey range, remaining lease and historical market conditions to estimate an expected transaction price.

Alongside the model, we built a comparables engine that surfaces similar historical transactions. This makes the prediction more interpretable: users can see the market evidence behind an estimate instead of receiving a black-box number.

We then aggregate transactions into time-series data to generate a 6-month market outlook for different HDB market segments.

A key part of our modelling approach is ensuring that historical features only use information that would have been available at that point in time, reducing the risk of data leakage and making our evaluation more realistic.

Challenges we ran into

One of our biggest challenges was that housing is extremely heterogeneous. Two flats in the same town can still differ substantially because of floor area, storey, lease remaining, flat model and market timing.

That meant a simple town-level average was not enough.

We also had to think carefully about data leakage. When calculating historical price benchmarks, it would be very easy to accidentally let future transactions influence the prediction of an earlier transaction. We therefore designed our features to respect transaction chronology and used time-aware model evaluation.

Another challenge was balancing accuracy with explainability. A buyer should not simply be told that a flat is worth "$X". We wanted FlatFair to show why the model reached that conclusion through comparable transactions and relevant market context.

Finally, this is a two-week sprint. There are many additional signals we could incorporate, so we had to prioritise building a focused, reliable end-to-end experience rather than trying to model every factor affecting property prices.

Accomplishments that we're proud of

We're especially proud that FlatFair evolved beyond being another housing dashboard.

We designed the product around an actual decision a buyer needs to make, connecting:

historical data → comparable flats → fair-value estimation → affordability → future outlook

We are also proud of building explainability into the product from the start. Our valuation model is complemented by real comparable transactions so users can understand the evidence behind an estimate.

Finally, we built the project as an end-to-end data product rather than a standalone model — from raw public data and transformation in Databricks through to analytics, machine learning and a user-facing experience.

What we learned

One of our biggest learnings was that a strong prediction is not necessarily the same as a useful product.

For a decision as important as buying a home, users need context, transparency and uncertainty, not just a model output.

We also learned how important time is when working with real-world market data. Features that appear harmless can leak future information into a model unless they are constructed carefully.

On the product side, we learned to start from the user's decision rather than from the available dataset. Instead of asking "What charts can we create from HDB data?", we started asking:

"What information would actually help someone decide whether to buy this flat?"

That question shaped FlatFair much more than any individual model or visualisation.

What's next for FlatFair

FlatFair's current scope focuses on building a strong foundation for HDB resale decision-making, but there is much more we would like to add.

Next, we would expand the platform with richer location-level features such as proximity to MRT stations, schools, amenities and employment centres.

We would also like to make affordability more personalised by incorporating financing scenarios, CPF usage, housing grants and different mortgage assumptions.

On the modelling side, we want to improve our forecasts with additional economic and housing-supply indicators and introduce prediction intervals so users can see a range of possible outcomes rather than a single forecast.

Ultimately, we see FlatFair becoming a personal housing intelligence layer — helping Singaporeans not only search for a home, but understand the financial decision behind it.