The Problem: When 50,000 Fans Hit the Gates at Once
Stadium bottlenecks aren't just annoying—they're dangerous. Static gate assignments fail when crowd dynamics shift. For PromptWar 2026, I wanted to solve this with code that actually thinks.
The Build: PromptWar Crowd Orchestration
Instead of rigid assignments, I built a 4-tier Human-Sensor Network that treats crowd management as a living system:
| Tier | Role |
|---|---|
| Host | Strategic oversight & emergency override |
| Volunteer | Ground-level sensors & mission execution |
| Attendee | Passive data source (location, velocity) |
| Service Provider | Resource allocation & logistics |
Technical Architecture
ElasticEntry Algorithm
The core routing engine calculates real-time "Friction Scores" for every entry point based on:
- Surge density (people/minute)
- Fan transit velocity
- Historical congestion patterns
Stack
- Backend: FastAPI + WebSockets for sub-second perimeter updates
- AI Layer: Gemini for predictive bottleneck detection (identifies trouble 10-15 min before it happens)
- Infra: Google Cloud Run—scales to zero, bursts to thousands
Live Demo
🔗 Try it on mobile
📂 GitHub repo
What I Learned
Combining real-time WebSocket state management with Gemini's predictive calls taught me where AI adds value vs. where deterministic logic wins. The sweet spot: AI for forecasting, traditional algorithms for immediate routing decisions.
Built for #PromptWar2026 with #GoogleCloud. Feedback welcome—how would you handle the load balancing differently?
