TransitIQ

GIF
Ranking neighborhoods by the cost of delayed transit investment, and showing how that cost compounds the longer a city waits

Inspiration

When you move countries as a student, public transport isn't a lifestyle choice — it's the only option. You learn every bus route, every transfer, every gap in the network. You budget carefully because a missed connection isn't just inconvenient — it's money you don't have, a shift you're late to, a healthcare appointment you couldn't reach.

We've lived this. Moving between countries, navigating cities where transit is excellent (Mumbai, Melbourne) and cities where it isn't, we saw firsthand how a well-connected city gives people real alternatives. In India and Australia, commuters choose public transport for work — not because they have no car, but because the system is reliable enough to trust. Mumbai's suburban rail alone carries over 7 million people daily, most of them working-class commuters for whom it is not a backup plan but the backbone of daily life. In cities where that investment never came, low-income residents don't have a choice at all.

That experience brought us to this question: what does it actually cost the people who depend on transit most when investment is delayed - not forever, just five years? What compounds in that time, and who pays for it?

What it does

TransitIQ is a decision-support dashboard for city planners that quantifies the cost of inaction — the compounding economic harm to low-income, transit-dependent workers when transit investment is postponed.

It answers one question, precisely:

"If this investment is delayed 5 years, here is the compounding cost to low-income workers — in measurable, specific terms."

Across 328 Franklin County (Columbus, OH) census tracts, TransitIQ surfaces:

Annual cost of inaction — lost wages, missed healthcare, environmental cost, education access, and forgone affordability, summed per tract
Compounding delay scenarios — what that cost becomes at 1, 3, and 5 years of delay, with differentiated growth rates per component
Per-resident and per-worker figures — $10.2K per transit-dependent worker over a 5-year delay
Priority triage — which tracts to fund first, ranked by combined hardship index and per-capita delay cost
Equity breakdown — which vulnerability segment absorbs disproportionate burden (High-need transit burden: $1,575/resident vs. $124/resident for the lowest-risk segment)
Responsible AI guardrails — named failure modes with who gets harmed and what design choices mitigate each

How we built it

Data Pipeline

We assembled six public datasets joined at the census tract level:

Dataset	Source	Use
ACS 5-Year Estimates (2019–2023)	US Census Bureau	Income, vehicle access, transit dependency
BLS OEWS 2024	Bureau of Labor Statistics	Columbus MSA wage rates
COTA GTFS 2025	Central Ohio Transit Authority	Bus stop coverage, route density
NHTSA FARS 2023	Federal Highway Administration	Safety warning overlay
LEHD LODES 2021	US Census / HUD	Jobs accessibility by tract
H+T Affordability Index	Center for Neighborhood Technology	Housing + transport cost burden

Cost Engine

The annual cost of inaction compounds five components forward at differentiated rates:

$$\text{Cumulative Delay Cost} = \sum_{y=1}^{N} \left( \sum_{c} \text{Component}_c \times (1 + r_c)^y \right)$$

Where wages compound at 3.5%, healthcare at 4.5%, and education at 1.2%. All assumptions are adjustable live via sidebar sliders.

Vulnerability Segmentation

K-means clustering (k=3) on the Transit Hardship Index segments tracts into High-need, Moderate hardship, and Lower-risk groups — used to drive the equity tab and priority triage.

Tech Stack

Python · Streamlit · Plotly · Pandas · GeoPandas · Scikit-learn · Nominatim (OpenStreetMap)

Challenges we ran into

The hardest part wasn't the code — it was the framing.

Our first version was a transit analytics dashboard. It had maps, charts, and hardship scores. But our mentor pointed out it was answering the wrong question. "Cost of inaction needs to be the core output — not just analytics." That reframe meant rebuilding the cost engine, restructuring every tab, and rethinking what a judge or planner should see first.

Other challenges:

Census tracts have no names — they're numbered administrative units. We reverse-geocoded 328 tract centroids via Nominatim to recover neighbourhood names for 166 tracts (Linden, Near East Side, Roseland, etc.)
Responsible AI beyond disclaimers — writing generic warnings was easy. Identifying the specific displacement paradox failure mode (investment → rising rents → residents priced out of the neighbourhood the tool recommended) and designing a concrete guardrail was not.
Live sensitivity — wiring sidebar sliders to recompute all five cost components and three delay scenarios in real time, without breaking the map, equity charts, or priority scores.

Accomplishments that we're proud of

The $10.2K number — reducing $978.8M in aggregate cost to a per-worker figure that makes the harm legible to a human being, not just a spreadsheet.
Honest Responsible AI — not a compliance checkbox, but three named failure modes (displacement paradox, suppressed data, borderline tract instability), each with a specific person harmed and a specific design response built into the tool.
Neighbourhood names from scratch — we had no name data. We geocoded every tract centroid and recovered real neighbourhood identifiers so planners see "Near East Side · Tract 54.10" instead of "39049005410."
End-to-end from raw Census data — ACS pulls, GTFS joins, cost engine, clustering, and a fully interactive dashboard, all built in the hackathon window.

What we learned

The most powerful number in this dashboard isn't the aggregate. It's the per-person number. $10,200 per transit-dependent worker is the figure that makes the cost of delay legible to a human being — and that's what changes the conversation in a city council meeting.

We also learned that data tools for public policy carry a different kind of responsibility than consumer products. When a dashboard influences where a city invests $25 million, the design choices — what you show first, what you call "high priority," whether you surface uncertainty — are not neutral.

Cities like Mumbai and Melbourne didn't become transit success stories because they had more money. They invested consistently, early, and with a clear view of who loses when they don't. TransitIQ is our attempt to give Columbus — and cities like it — the same clarity.

What's next for TransitIQ

Per-tract data quality flagging — surface which tracts have suppressed or imputed Census values so planners can weight those recommendations accordingly
Soft clustering — replace hard K-means boundaries with fuzzy c-means to show borderline tract confidence
Expand beyond Columbus — extend the cost engine to additional Ohio MSAs and eventually other US metros
Community review integration — flag tracts for direct resident input before funding decisions reach council