RescuePathAI – مسار الإنقاذ الذكي
منصة ذكية لإدارة الكوارث وإنقاذ الأرواح. An AI-powered platform that helps residents stay safe during disasters and missile threats, and helps emergency teams prioritise and respond faster.
Live demo: rescuepath-ai.vercel.app (resident app) · /dashboard (operations dashboard)
⚠️ Prototype. All alerts, zones, shelters, routes and incidents are simulated demo data for Riyadh. This is not a real emergency service. In a real emergency call 911.
Features
Resident app (/)
- Interactive risk map (Riyadh, OpenStreetMap): affected zones, incidents, closed roads, shelters and assembly points.
- Scenario simulation: missile alert (North Riyadh), flash flood (Wadi Hanifah), industrial fire (Al Sulay).
- Early warning: residents inside the affected zone get the alert; residents outside are told to avoid the area.
- Shelter vs. evacuation guidance: missile alerts route you on foot to the nearest approved shelter, or tell you to shelter in place when none is within a 5-minute walk; floods and fires route you to the nearest safe assembly point outside the zone.
- Hazard-specific safety steps, including simple wording for children and older adults, plus emergency numbers.
- Four ways to report: take a photo, upload a photo, record a voice note, or write a description.
- Photo analysis: Claude looks at the incident photo and suggests the hazard, severity and teams; without an API key a clearly labelled simulated analysis is shown instead.
- Instant SOS: press and hold to send your location as a critical report, for people who can't type or talk.
- People affected: −/+ counters for adults, children, older adults, people with disabilities, pregnant women and people trapped, with a live total. Vulnerable groups are shared only with consent.
- Simulated voice notes: real recording in the browser; the transcript is a sample the resident can edit.
- Live response status: new → responding → rescued, with the teams on the way and their ETA.
- Danger zones and alternative routes: zones and closed roads in red with warning markers; routes that cross another danger zone or a closed road are replaced by an alternative (simulated where the demo data has no real road closure).
- Simultaneous scenarios: run the missile, flood and fire scenarios at once; each brings its own incoming reports.
- Offline mode: guidance and numbers are bundled in the app; SOS reports made without a connection are saved, clearly marked "not sent yet", and sent automatically on reconnect.
Operations dashboard (/dashboard)
- AI triage: each report is classified by hazard type, severity (🟢 low → 🔴 critical), priority score and needed team, with reasoning in Arabic and English.
- Priority queue: ordered by severity, people affected and vulnerability, not arrival time.
- Human review: dispatchers confirm the AI assessment or correct the severity; corrections re-rank the queue.
- Required resources: number and type of teams plus equipment, from the hazard, severity and people affected.
- Multi-team AI dispatch: suggests the nearest available team for each needed type, serving the most urgent cases first so a lower-priority case never takes a team a critical one needs. Falls back to a capable substitute team when none of the right type is free. The dispatcher approves every assignment, or all at once.
- Critical alerts: a red banner (and a beep) when a new critical report arrives.
- Report summary: hazard, area, people affected and resources needed in one paragraph.
- Duplicate detection: same hazard within 300 m and 30 minutes is flagged; merge it into the original or dismiss.
- Reassessment: update the people affected or what changed; the case is re-rated and re-ranked, with a history.
- Heat map: risk by area from open reports and active zones.
- Response reports (
/dashboard/reports): reports handled, response times, outcomes, and team utilisation.
- Team tracking: team status (available / dispatched / on scene) on the map, and case status updates through to closure with an outcome.
Both views are Arabic-first (RTL) with an English toggle.
Getting started
npm install
cp .env.example .env.local # optional: add ANTHROPIC_API_KEY for AI triage
npm run dev
Open http://localhost:3000, then use the Simulation panel to start a scenario, press "Move me into the zone", and try sending an SOS. You can also drag the blue marker to change your position.
Open http://localhost:3000/dashboard in a second window to see the report arrive, review it and dispatch a team, then watch the resident's "My reports" tab update. To send reports from a phone, open the dev server's Network URL on the same Wi-Fi. Reset demo on the dashboard restores the sample cases.
Cases and teams are kept in the server's memory, so they reset when the server restarts, and the prototype assumes a single server instance. A real deployment needs a database and authentication for the dashboard.
AI triage
POST /api/reports sends each new report to Claude (claude-opus-5-5, low effort, structured output) when Anthropic credentials are available. If there are no credentials, or the call fails, the app falls back to a keyword- and rule-based classifier (src/lib/triage/rules.ts), which also triages the sample cases, so an SOS is never blocked by the AI service. Set TRIAGE_MODE=rules to skip the AI entirely.
Deployment
The live demo is hosted on Vercel, and every push to main deploys automatically. Add ANTHROPIC_API_KEY under the Vercel project's environment variables to enable Claude triage there; without it the demo uses the rule-based classifier. Because cases live in server memory, the hosted demo can reset after it has been idle.
Project structure
src/
app/
page.tsx # resident app
dashboard/page.tsx # operations dashboard
api/reports/ # residents: submit SOS (triaged), poll response status
api/ops/ # dashboard: live state, case actions, scenario, demo reset
components/ # resident UI, dashboard/, map/
lib/
data/ # scenarios, shelters, guidance, response teams
triage/ # Claude triage + rule-based fallback
dispatch.ts # priority ordering and team suggestions
server/ops-store.ts # in-memory case and team store
geo.ts, routing.ts # zone checks, nearest safe place, OSRM routing
store.tsx, i18n.tsx # resident state and Arabic/English strings
Tech: Next.js 16, React 19, TypeScript, Tailwind CSS 4, Lucide icons, Leaflet with OpenStreetMap tiles, OSRM routing (routing.openstreetmap.de), and the Anthropic SDK.
Roadmap
- Persistent storage (database) and dispatcher sign-in.
- Route-aware dispatch that avoids closed roads.
- Family safety check-in: "I'm safe" / "I need help" for family members.
- Predictive flood-risk layer from rainfall, terrain and past incidents. Missile threats rely on official alerts only and are never predicted.
- Installable PWA with a service worker for full offline support.
