Kaki-Lator

Inspiration

Hawker centre culture is pivotal to Singaporeans' way of life and has shaped our cuisine in more ways than one. As a couple of foodies, food is often the subject of our hangouts, and this problem statement resonated with our mutual agreement that hawker centre access isn't as easy as it ought to be. We believe that being able to create a product that allows planners to generate insights and suggestions on future plans for Singapore's hawker landscape would be helpful in sustaining this aspect of Singaporean tradition that we all know, love and appreciate.

What it does

Our product labels each Singapore planning area by how its hawker access fails, on two different metrics: how thin the supply of taker centres are (number of hawker centres per resident), and isolation (next closest hawker centre).

It has a closure what-if analysis: if a hawker centre were to be closed today, what is the nearest hawker centre that residents can head to? Is there a fallback within 1km? How long does the closure last and how many people does this closure affect?

It is built for NEA planners, and town council estate managers. They can use our product to schedule for long repairs (so that overlaps do not occur between neighbouring hawker centres), flag areas for interim support (planning new coffee shops in place of a long closure for isolated hawker centres), and see which gaps remain after new hawker centres open.

How we built it

Done: we analysed the NEA hawker centre and closure files, Census 2020 population by planning area, and the URA planning-area boundaries (all from data.gov.sg).

Planned: a bronze, silver, gold (medallion) pipeline on Delta tables in Databricks Free Edition. Bronze holds raw copies, silver the cleaned and joined data, and gold three tables for area, centre and closure.

Planned: the classification and what-if are rule-based SQL, so that every finding can be explained.

Planned: an AI/BI dashboard plus a Genie space for plain-English questions. Genie writes the SQL and the numbers come from the SQL, not the model.

Challenges we ran into

The closure data is patchy: 26 slots have "TBC" dates and were excluded, and it has no permanent closures. A permanent-versus-temporary trend can't be shown.

Additionally, the data has no footfall or capacity figures, and distances are straight-line, so the what-if analysis is a modelled change in access, not measured behaviour.

The population and hawker centres data come from different years (Census 2020 against 2026 centres), and census values are rounded to 10 with some suppressed.

Accomplishments that we're proud of

We are proud that we found a distinctive angle to frame our problem statement (and consequently solutions) that holds up in our data, rather than a generic density dashboard.

What we learned

We were originally going to argue for lack of accessibility in neighbourhoods with a distinctively older demographic, but it turns out that hawker centre distribution overwhelmingly favours older neighbourhoods, and the real distribution gap and isolated hawker centre problem is found in younger neighbourhoods.

Closure clashes are rare. Only 6 centres ever had all neighbours within 1 km shut on the same day, for 5 days at most. The real exposure is isolated centres, which carry 27% of 2026 closure centre-days (415 of 1,533).

The planned centres barely close the gap. Yew Tee lifts Choa Chu Kang only to about 0.05 hawker centres per 10,000 residents, and Tampines North lifts Tampines to about 0.12.

What's next for Kaki-lator

Sprint (12 to 26 Oct): week 1 covers ingestion, cleaning, placing each centre in its area and the gold tables. Week 2 covers the closure what-if, dashboard, Genie and rehearsal.

Beyond: add capacity data and walking-route distances, refresh with newer population data, and test the what-if analysis with real planners, since its value to them is still an assumption.

Built With

• ai-dashboards
• bi-dashboards
• data.gov.sg
• databricks
• databricks-sql
• delta-lake
• genie
• lakeflow
• python
• sql