Inspiration
Houston already has heat maps. What operations teams still need is a defensible answer to a harder question: where should the next cooling dollar go when event activity, neighborhood vulnerability, existing resources, cost, time, and uncertainty all compete?
What it does
HeatSignal is an evidence-aware planning workspace for Houston's 2026 World Cup footprint and the investments that should last beyond it. A planner switches between Houston Stadium and the downtown FIFA Fan Festival, selects heat conditions, budget, delivery horizon, and equity preference, and receives:
- a switchable priority, heat, vulnerability, and cooling-gap map;
- an optimized intervention portfolio that never exceeds budget;
- cost ranges, modeled daily capacity, owners, lead times, risks, rationale, and prerequisites;
- scenario saving, baseline comparison, CSV export, and a print-ready action brief;
- an implementation curve from T−12 weeks to event day or through Year 5;
- a data-quality view with official links, vintages, hashes, spatial-join rules, assumptions, and privacy boundaries.
How we built it
Python and DuckDB refresh and spatially aggregate governed public evidence into Houston's 88 named Super Neighborhoods. The model area-weights 2022 CDC/ATSDR SVI and an NCAR summer-heat surface, assigns mapped cooling resources by point-in-polygon containment, and creates an event-activity proxy from daytime-population density and distance to each event footprint.
An exact grouped-knapsack optimizer selects at most one intervention per zone and never exceeds budget. It precomputes 800 reproducible scenarios across two sites, two heat conditions, two horizons, five equity weights, and twenty budgets. Invariant intervention metadata is published once in a candidate catalog; compact scenarios reference candidates by ID and are strictly hydrated and validated in the browser. React, TypeScript, MapLibre GL JS, and SVG turn those artifacts into a fast static application.
Data integrity
The hardest work was avoiding a persuasive but false join. Organizer visits have no coordinates and combine Dallas with Houston; weather is station/day; only organizer UHI is spatial. We keep those grains separate and private. No private record is copied into the browser.
The public experience uses traceable government and government-hosted snapshots instead. It clearly labels the NCAR heat and resource layers as 2014 evidence needing a current operational refresh.
AI and Codex
AI was used as a development and analytical copilot, not as an unreviewed runtime decision-maker. It helped audit source grains, challenge unsafe joins, structure the intervention model, test assumptions, identify performance and accessibility gaps, and translate technical evidence into judge-facing documentation. The shipped prioritization and allocation logic is deterministic and reproducible.
Codex supported the complete product cycle: competition analysis, public-source review, confidentiality boundaries, Python/DuckDB pipeline work, optimization, React/TypeScript implementation, MapLibre interactions, responsive design, accessibility, release configuration, browser automation, video generation, and submission preparation.
Accomplishments
- 88 real, named operational planning zones.
- 800 exact, budget-feasible scenarios and 1,018 intervention candidates.
- Two official event footprints and 201 mapped cooling resources.
- Six intervention types with owners, prerequisites, timing, cost, capacity, risk, and robustness.
- Match-day and five-year impact pathways.
- Reproducible source snapshots with manifests and SHA-256 hashes.
- 16 Python tests, 10 TypeScript tests, strict type-checking, production build checks, and desktop/mobile browser verification.
- A responsive decision workflow rather than a gallery of charts.
Results
In the default Houston Stadium scenario—typical heat, match-day operations, $250,000, and 50% equity—the optimizer funds two zones, models 1,600 people/day of service capacity, and covers 2.0% of total weighted citywide need (1.4–2.3% sensitivity).
An extreme-heat, $500,000, 75%-equity scenario spends $475,000 across five zones, models 2,950 people/day, and covers 5.0% (3.9–5.8%).
These figures are transparent planning assumptions and normalized coverage metrics. They are not predicted illness reductions, approved sites, procurement estimates, or claims about observed World Cup visitors.
What is next
Before operations, Houston partners should refresh the 2014 heat/resource layers, replace cost and capacity assumptions with procurement evidence, field-check shortlisted locations, co-design with communities, and establish pre-event baselines. The same model can then support festivals, conventions, heat emergencies, and long-term shade capital planning.
Built With
- codex
- duckdb-spatial
- ffmpeg
- maplibre-gl-js
- playwright
- python
- react
- typescript
- vite
- vitest