The Sandwich

Inspiration

Walk through any HDB estate at 2pm and watch who is actually moving. Not the cycling-path commuter the planners pictured. A grandmother pushing a stroller. A shift worker catching a weekday clinic appointment. A parent sprinting to pick up a sick kid from childcare. The cycling path they are supposed to use is empty, because at 32°C with a dependent in tow it is not inclusive. It is unusable.

That gap, between infrastructure that is counted as inclusive and infrastructure that actually works, is the whole project.

What it does

A heat-adjusted accessibility engine. It scores every planning area in Singapore by the share of daylight hours in which a sandwich-generation care circuit (childcare, AAC or polyclinic, home) is physically completable without a car.

The output is a three-view AI/BI Dashboard: a national map of planning areas colour-coded by heat-adjusted accessibility, a ranked gap list of areas where the care circuit fails most hours of the day, and a Master Plan 2025 overlay testing whether planned cycling additions close the gap or miss it.

The user is an LTA or URA planner deciding where to prioritise new active-mobility infrastructure. The score reframes that decision from static distance coverage to time-conditional usability.

How we built it

Hypothesis-first, dashboard-second. The thesis, that Singapore's active-mobility infrastructure is only inclusive at the edges of the day, drives the pipeline. Not the other way around.

• Notebook ingestion of cycling, population, facility and real-time temperature data from data.gov.sg into Delta Lake tables.
• PySpark geospatial joins to compute distance from each planning area's population centroid to its nearest childcare centre, AAC and polyclinic.
• A composite heat-adjusted accessibility score, tracked in MLflow, that weights raw distance by hourly temperature and historical rainfall.
• An AI/BI Dashboard with the three views above.

Unity Catalog governance and a Genie natural-language layer are stretch goals, not MVP.

Challenges we ran into

1. I am a first-year HR student entering a technical hackathon. Problem framing, policy critique and storytelling are the toolkit. The Databricks stack is new territory. The bet is that AI-assisted coding plus a sharply defined thesis beats brute-force engineering with a vague question.
2. Solo team, two weeks, interning full-time. Scope discipline is the whole game. One slice of the problem, one score, one dashboard, then stretch only if MVP lands early.
3. The data proves correlation, not causation. Heat makes cycling paths less usable, but delivery apps, car dependency and estate design all muddy the picture. The deliverable has to be honest about what it does and does not show.

Accomplishments that we're proud of

Catching a thesis-killing dataset mismatch before anchoring the pitch to it. One dataset listed in the brief as "Hawker Centre Closures" turned out to be the quarterly cleaning schedule, not permanent closures. We burned a day pivoting, but we did not submit a pitch built on a dataset that could not support it.

Reframing the sandwich generation as a function instead of a demographic. A working parent, a grandmother and a shift worker can all run the same care circuit on the same day. That reframe is what makes the heat-adjusted lens defensible against the obvious objection ("don't most people commute at 7am?").

What we learned

• Verify every dataset before anchoring a thesis to it. The brief's description is a starting point, not ground truth.
• "Inclusive mobility" is a policy term that quietly assumes an idealised user. The idealised user is not the person most reliant on the infrastructure working.
• A sharp thesis beats a broad one. Testing one specific claim (the care circuit fails under heat) is more actionable than mapping accessibility in the abstract.

What's next for The Sandwich

If shortlisted, build the MVP in the two-week sprint window: end-to-end Delta pipeline, PySpark geospatial scoring, the three-view dashboard. Then push the stretch (Unity Catalog, Genie) if the MVP lands early.

Longer term, the real ambition is not one dashboard. It is a reusable inclusivity metric that LTA and URA can run against every future active-mobility proposal. Climate-adjusted accessibility becomes the default stress test. Infrastructure that only works before 10am and after 6pm stops being counted as inclusive by default.

Built With

• genie
• sql