Urban heat intelligence that moves from detection to field-ready action.
Built for FortyGuard Global AI Hackathon ’26
Primary track: Resilient Cities & Infrastructure
A temperature surface tells a city where heat is. It does not tell a resilience team which blocks deserve attention first, why one zone ranks above another, what evidence supports that decision, or what should happen next.
KAIRO closes that operational gap with one legible decision chain:
Detect → Analyze → Explain → Prioritize → Act
It combines verified FortyGuard temperature output with explicitly labeled contextual evidence, transparent scoring, route comparison, intervention screening, and concise assessment briefs. The result is a decision-support workspace designed for municipal teams, not another decorative heat map.
|
|
| Desktop operations workspace | Responsive field view |
- A completed FortyGuard Phoenix TCM activity returns 1,055 normalized temperature tiles.
- The live map renders only values actually returned by FortyGuard, with no hidden fallback or silent scenario substitution.
- Every live tile can be inspected for its exact TCM value, with a clearly labeled relative scale for narrow observed ranges.
- The Canopy control opens the labeled City of Phoenix scenario layer instead of presenting a disabled or ambiguous action.
- Credentials stay server-side; the browser never receives the API key or raw upstream payload.
- API responses pass schema validation before normalization and visualization.
| Stage | What KAIRO adds |
|---|---|
| Detect | Hyperlocal thermal field and peak/mean/minimum TCM summaries |
| Analyze | Robust anomaly detection, exposure scoring, persistence, and correlations |
| Explain | Visible evidence provenance and human-readable drivers for every priority |
| Prioritize | Deterministic hotspot ranking across 30 Phoenix analysis zones |
| Act | Cooler-route comparison, intervention screening, and downloadable field briefs |
- A decision layer, not only a visualization. Every signal continues into a priority, explanation, and next assessment step.
- Evidence boundaries are visible. Observed, derived, public-data, and modeled scenario fields remain separate and labeled.
- Explainability is part of the product. Exposure weights, median/MAD anomalies, Pearson/Spearman associations, and ranking logic are documented instead of hidden behind a score.
- The Analyst is bounded. It routes questions to deterministic local analytical tools; it does not invent measurements or depend on an unrestricted runtime LLM.
- Live and scenario modes never blur. The verified FortyGuard result proves integration; the Phoenix Scenario demonstrates the complete operational workflow.
flowchart LR
FG[FortyGuard Temperature API] -->|server-only request| AD[Validated API adapter]
AD -->|normalized TCM tiles| LIVE[Verified Live Mode]
CITY[City of Phoenix canopy sample] --> SC[Phoenix Scenario]
SC --> ENG[Deterministic intelligence engine]
LIVE --> UI[Decision workspace]
ENG --> UI
UI --> OUT[Hotspots · routes · scenarios · briefs]
The production path uses a server-only POST /v1/heatmap request, bounded polling through GET /v1/status/{activity_id}, Zod validation, geometry normalization, and a responsive Canvas renderer. The analytical layer is pure TypeScript and independently tested.
- Application: Next.js 16 App Router, React 18, strict TypeScript
- Interface: Tailwind CSS, local shadcn-style primitives, Lucide icons, Recharts
- Geospatial: GeoJSON, Turf helpers, responsive Canvas choropleth
- Validation: Zod at external-data boundaries
- Testing: Vitest for analytical, routing, scoring, and API-normalization logic
- Deployment: Vercel with encrypted server-side environment configuration
npm installCopy .env.example to .env.local, then add a FortyGuard key:
FORTYGUARD_API_KEY=your_server_side_keyFORTYGUARD_BASE_URL is optional. Never expose the key through a NEXT_PUBLIC_ variable or commit .env.local.
npm run devOpen http://localhost:3000. The Phoenix Scenario works without a key; verified Live Mode is enabled only after a valid FortyGuard response completes.
npm run lint
npm run check
npm run buildCurrent release gate:
- ESLint: clean
- Vitest: 25/25 passing
- Production build: successful, strict TypeScript included
- Responsive verification: desktop and 375 px mobile, no horizontal overflow
- Interaction verification: Canopy transition, live-tile inspection, tablet, and mobile-landscape checks
- Production checks: public routes return 200; CSP and HSTS are enabled
| Document | Purpose |
|---|---|
| Submission write-up | Concise project narrative for judges |
| Three-minute demo script | Timed presentation flow |
| Data methodology | Provenance, transformations, and analytical boundaries |
| FortyGuard integration | Authentication, polling, validation, and failure behavior |
| Architecture | Components, data flow, security, and deployment |
| Hackathon evaluation | Evidence against the official judging rubric |
KAIRO is a hackathon decision-support prototype. It is not a certified safety, health, navigation, engineering, or construction tool. Scenario reductions are screening estimates, not forecasts or guaranteed outcomes. Correlation does not imply causation. Thirty canopy percentages come from the City of Phoenix 2022 Shade Study sample; other contextual scenario fields are deterministic and labeled. Current FortyGuard API coverage is limited to the United States.
Built by Marwan Abdelghaffar and Basel Ziada for FortyGuard Global AI Hackathon ’26.
See the heat. Explain the risk. Prioritize the next move.