Inspiration The idea hit during the 2022 World Cup. 80,000 people in one stadium and the only tools we had were static signs and 20-minute bathroom lines. Vendors were guessing demand, fans were missing goals, and operations was flying blind.
For FIFA 2026, we wanted to build the MARACAÑÃ STADIUM COMMAND DECK — a real-time nervous system for the venue. One screen that lets fans find beer in 3 minutes, vendors prep for surges before they happen, and ops see density per section.
The name StadiumSurgeSync comes from the core loop: detect a surge, then sync every stakeholder to respond.
What it does StadiumSurgeSync is a real-time AI ops dashboard for stadium events.
For Fans: Natural language Fan Deck. Ask “beer in Section East, fast” and get the exact stand, live wait time, and a route. Ask “bathroom <5 min” and avoid halftime chaos.
For Vendors: Vendor Kiosk gets surge alerts + prep recs. Example: “Density in West is 56%. Prep 80 refreshments, 100 tacos.”
For Ops: Live Grid Activity Map shows density per section. Multi-agent loop writes events to MongoDB when thresholds are crossed.
Brazil vs France, 74’, 1–1. West section hits 56% density → Fan gets routed to 3-min beer line → Vendor gets prep alert.
How we built it Stack
- Frontend: React Native + Expo for cross-platform mobile
- Backend: http://Node.js + Express
- AI: Gemini 3 Flash for Fan Deck, Vendor Kiosk, Fantasy Coach agents
- Database: MongoDB Atlas with Triggers + Change Streams for real-time sync
- Deployment: In-Memory Cluster Sandbox for demo speed
Architecture
- Telemetry: Simulated crowd sensors push section density to
live_eventscollection - Threshold Logic: MongoDB Trigger fires when density > 55% \text{Alert} = \begin{cases} \text{True} & \text{if } \text{Density}_{section} > 0.55 \ \text{False} & \text{otherwise} \end{cases}
- Multi-Agent Loop: Gemini agents read the event, generate fan guidance + vendor prep instructions
- Sync: Change Streams push updates to all clients <200ms. Fan sees 3-min wait. Vendor sees prep rec.
Data Model
live_events — 50 docs of real-time matches, minutes, scores, attendance
vendors — 8 docs with location, capacity, catalog, live wait times
Challenges we ran into
- Noisy demo environment: Recorded 3 separate clips in a loud room. Audio was unusable. Fixed it by focusing on screen record + captions. Added pinned comment to clarify “StadiumSurgeSync”.
- Latency vs accuracy: Real-time means nothing if the beer line moved while you walked there. We solved it with MongoDB Change Streams + optimistic UI. Updates hit the app in ∼180ms.
- Hallucinated vendors: Early Gemini responses invented stands that didn’t exist. We grounded all prompts with actual
vendorscollection data. Now it only routes to Carioca Brew Stand 4E because it’s in the DB. - Threshold tuning: 50% density was too noisy, 60% was too late. Landed on 55% after testing: \text{Optimal Threshold} = \mu_{density} + 1.2\sigma \approx 0.55 That catches surges 2–3 minutes before queues form.
- YouTube UI hell: Couldn’t find Share button to get the link. Learned the “open video first, then Share” flow. Submitting a hackathon teaches you everything.
What we learned
- Real-time is a UX problem, not just a DB problem. 200ms updates are useless if the fan doesn’t trust the 3-min wait time. We added “Updated 12s ago” to build trust.
- Multi-agent > single chatbot. Fan, Vendor, and Fantasy agents need different context. A beer query shouldn’t see fantasy stats.
- MongoDB Triggers are OP for ops. Writing business logic at the data layer means every client stays in sync without polling.
- Ship > perfect. Our first demo had wrong captions and no audio. We still shipped because the core loop worked: sense → alert → action.
What’s next
- Real sensors: Integrate with WiFi access point counts + POS data instead of simulation
- Predictive surges: Use historical minute + score data to predict 70’ bathroom rush before it happens
- FIFA 2026 pilot: Talking to 2 host committees about Maracanã and MetLife trials
Built for fans who want to watch the match, not wait in line.