An AI-powered command center that unifies crowd intelligence, live navigation, sustainability tracking, and a multilingual fan assistant into a single real-time operating system for large-scale stadium events — built for FIFA World Cup 2026.
Smart Infrastructure & Public Safety — AI for Large-Scale Event Operations
We targeted the intersection of crowd safety, real-time operations, and fan experience at mega-events. Stadiums hosting 80,000+ attendees face a recurring set of problems that existing systems handle in silos — CCTV monitoring, transport logistics, crowd density, sustainability reporting, and fan support are usually separate tools run by separate teams. Our vertical is unifying these into one AI-driven control layer, serving two distinct user groups from a single codebase:
- Fans — a public-facing experience for navigation, transport, and multilingual support.
- Operators — a control tower for live monitoring, anomaly detection, and rapid incident response.
We chose this vertical because it combines measurable operational impact (faster evacuation routing, reduced bottlenecks, lower incident response time) with a genuinely useful AI layer (the assistant and anomaly detection aren't cosmetic — they answer real queries and flag real patterns), rather than being a UI wrapper around static data.
Rather than building separate apps, we built a single React application with two rendering modes driven by app state:
currentMode === "landing"→ Fan Mode: public dashboard, live scores, chat assistant, transport/sustainability widgets.currentMode === "ops"→ Control Tower Mode: operator dashboard with analytics, CCTV AI vision, emergency ops, and the Jarvis-style ops assistant.
This kept the codebase DRY — both modes share the same data hub, the same design system, and the same AI backend, just with different surfaced panels and permission levels.
We split the system into three logical layers so each could be built, tested, and reasoned about independently:
- Data Layer (
worldcupApi.ts) — a single source of truth for match data, deduplicated and normalized (order-insensitive team-pair matching) so the same match reported by different feeds never renders twice. - Intelligence Layer (
gemini.ts,server.ts) — a multi-model AI fallback chain plus a deterministic "synthetic brain" fallback, so the assistant never goes silent even if the primary model is rate-limited. - Presentation Layer (component tree) — a card-and-panel grid where every feature (Crowd Intelligence, Transport Hub, CCTV AI Vision, etc.) is a self-contained, independently clickable module, so operators can jump straight to the panel they need instead of scrolling a monolithic dashboard.
Real stadium operations software has to survive two failure modes: data conflicts (duplicate/contradictory feeds) and AI service outages (rate limits, downtime) — both are common in live event environments. We designed the deduplication engine and the multi-model fallback specifically to make the system resilient to both, rather than assuming a perfect network/API environment.
- User lands on the hero page and sees live match state, capacity, and AI confidence metrics pulled from the data hub.
- They can query the FIFAiq Assistant in natural language (10+ languages auto-detected) for things like nearest gate, food options, or transport — answered either by a live Gemini call or, if unavailable, an instant local knowledge-base response so there's no perceptible downtime.
- Cards for Transport, Sustainability, and Navigation surface real-time-style widgets (shuttle times, waste diversion %, sector wayfinding).
- Operator clicks Enter Control Tower, switching the app into Ops mode.
- The Crowd Intelligence panel shows 10-sector density monitoring with CCTV overlays and AI anomaly flags.
- Related panels — CCTV AI Vision, Analytics Engine, Emergency Ops — are each one click away from the same grid, each opening its own focused view rather than a single overloaded screen.
- Emergency Ops can trigger dynamic evacuation routing and signage overrides directly from detected anomalies, closing the loop between detection and response.
/api/worldcup,/api/gemini,/api/gemini-ops,/api/jarvis, and/api/anomalyare served from a single Express backend.- Gemini calls cascade through multiple models (
gemini-2.5-flash → gemini-1.5-flash → gemini-2.0-flash → gemini-1.5-pro) before falling back to a deterministic local response engine — guaranteeing the assistant always answers. - Rate limiting and CSP headers protect the API layer from abuse.
Every interactive element — including all 9 feature cards — is a real, keyboard-operable control with visible focus states and descriptive labels, so operators and fans using assistive technology can navigate the same functionality as mouse/touch users.
- Live data availability: Where a live official feed wasn't accessible during development, we used authoritative historical/reference data (e.g.
openfootball/worldcup.json) as a stand-in, structured so it can be swapped for a real live feed without changing the consuming components. - CCTV/anomaly detection: Camera feeds and anomaly flags are simulated for demo purposes; the architecture (panel, alert pipeline, response actions) is built to plug into a real computer-vision backend without UI changes.
- Single-venue context: The demo assumes one active stadium/session at a time. Multi-venue support (concurrent matches across 16 stadiums) is architecturally possible via the existing stadium knowledge base but isn't exposed in the current UI.
- User roles: We assume two roles only — general fan and stadium operator — with no granular permission tiers (e.g. security vs. medical vs. transport staff) within Ops mode yet.
- Network conditions: The AI fallback chain assumes intermittent, not total, connectivity loss. Full offline operation is out of scope for this build.
- Language coverage: Multilingual support covers the 10 languages implemented in the knowledge base; broader language coverage would require extending the synthetic response engine's dictionaries.
Vite + React 19 + TypeScript · Tailwind CSS v4 · Three.js / React Three Fiber · Framer Motion · Express + Node.js · Gemini API · Recharts · Jest