PitchOps — Smart Stadiums & Tournament Operations (FIFA World Cup 2026)

Google Prompt Wars · Challenge 04 — Smart Stadiums & Tournament Operations

A GenAI-powered, multi-role assistant that optimizes stadium operations and elevates the FIFA World Cup 2026 experience for fans, volunteers/staff, and ops control rooms. Powered by Google Gemini via the Lovable AI Gateway, backed by Lovable Cloud (Postgres + Auth), and designed for Google Maps Platform + Google Calendar integration.

⚽ PitchOps — Smart Stadiums & Tournament Operations (FIFA World Cup 2026)

🚀 Google Prompt Wars • Challenge 04

A GenAI-powered platform to optimize stadium operations and enhance the FIFA World Cup 2026 experience through intelligent, real-time assistance.

🌍 About The Project

PitchOps is a Generative AI-powered Smart Stadium Operations Platform built for Google Prompt Wars Challenge 04.

The platform assists:

👥 Fans
🦺 Volunteers
🏟 Stadium Staff
🚓 Security Teams
📊 Operations Control Rooms

by providing intelligent, multilingual, real-time assistance using Google Gemini AI.

The goal is to improve:

Smart Stadium Navigation
Crowd Management
Emergency Response
Accessibility
Transportation Guidance
Sustainability
Volunteer Assistance
Operations Intelligence
Real-time Decision Support

for the FIFA World Cup 2026.

🏆 PromptWars Virtual Participation

Build • Deploy • Win From Anywhere

This repository is actively being developed as part of PromptWars: Virtual India.

Event	Details
Organizer	Google for Developers × Hack2Skill
Event	PromptWars Virtual
Region	🇮🇳 India Only
Format	Bi-weekly Virtual Hackathon
Development Style	Prompt-first Engineering
AI Platform	Google Gemini
Workflow	Google Antigravity

Official Event

https://promptwars.in/promptwarsVirtual.html

📅 How The 14-Day Cycle Works

Phase	Timeline	Description
🚀 Challenge Release	Day 1 (Monday)	A new real-world AI challenge is released.
💻 Building Phase	Day 1 – Day 12	Build a production-ready solution using Google Gemini and prompt-first workflows.
📤 Submission Phase	By Day 13	Submit source code, live demo, blog and LinkedIn Build-in-Public post.
🏅 Evaluation	Day 14	Experts review projects and update the leaderboard.

Live URL: https://pitch-ops.vercel.app/

1. Approach & Logic

The problem statement asks for one solution that improves navigation, crowd management, accessibility, transport, sustainability, multilingual assistance, and real-time decision support during matchday. Rather than building three siloed apps, PitchOps ships one app with three role-scoped experiences:

Role	What they get
Fan	Multilingual matchday companion — chat in 7 languages, live gate/section metrics, wheelchair-accessible amenity routing, transit link to the venue, Google Calendar match-add.
Volunteer / Staff	Incident reporter with automatic Gemini classification (kind + severity), SOP lookup, live Kanban triage board (new → dispatched → in-progress → resolved).
Ops Control	Live crowd heatmap, streaming Gemini Ops Brief with prioritized recommendations grounded in current metrics + open incidents, multilingual PA broadcast composer with tone control, sustainability KPI panel.

Why "one app, three roles"?

Data flows across roles. A fan-facing amenity route reuses the same section metrics an ops team acts on. A staff-reported incident feeds the same Gemini brief ops reads. Splitting the app would duplicate logic and hide these dependencies.
Realistic tournament UX. A single volunteer might act as staff during ingress and consume the fan view during a match — one auth session, one role switcher.
Judge-ready in one screen. Every rubric track is demonstrable from the same login.

GenAI logic

Model: google/gemini-3-flash-preview for everything (chat, classification, structured ops brief, multilingual translation). Fast, cheap, and multimodal-capable if we later add photo attachments to incidents.
Streaming chat uses the Vercel AI SDK (streamText + toUIMessageStreamResponse) so responses feel live and support message.parts rendering.
Grounding. Fan chat receives a compact VENUE CONTEXT block on every request (venue name, seat, live per-section occupancy + gate waits). Ops brief pulls venue + metrics + incidents at generation time. The system prompt forbids invented numbers.
Role-scoped system prompts. Fan = warm & simple. Staff = procedural & terse. Ops = analytical & metric-citing.
Structured output (Ops brief, incident classification, broadcast translation) uses prompt-driven JSON parsing with a safe fallback — deliberately avoiding heavy schema constraints that some models reject.

2. How the solution works

End-to-end matchday flow

                      ┌────────────────────────────┐
                      │  Landing / Auth (/, /auth) │
                      └──────────────┬─────────────┘
                                     │  Email + Google OAuth
                                     ▼
              ┌──────────────────────────────────────────┐
              │  /_authenticated  (role-gated shell)     │
              │  role switcher + sign-out                │
              └──┬──────────────┬─────────────────┬──────┘
                 │              │                 │
                 ▼              ▼                 ▼
             /fan           /staff              /ops
        fan companion   incident console   command center
                 │              │                 │
                 └──────┬───────┴────────┬────────┘
                        ▼                ▼
              /api/chat (stream)   server functions
          Gemini streaming chat   ops.functions.ts
                                  incidents.functions.ts
                                  broadcast.functions.ts
                        └────────┬────────┘
                                 ▼
                    Lovable AI Gateway → google/gemini-3-flash-preview
                                 +
                    Lovable Cloud (Postgres + Auth + RLS + Realtime)
                                 +
                    /api/public/tick (matchday simulator)

Data model

Every table has RLS + explicit GRANTs. Roles live in a separate user_roles table with a has_role(user_id, role) SECURITY DEFINER helper — never on the profile.

Table	Purpose
`profiles`	Display name, preferred language, accessibility prefs.
`user_roles` (+ `app_role` enum: `fan`/`volunteer`/`ops`)	Role assignments. Auto-`fan` on sign-up via trigger.
`venues`, `sections`	Seeded with MetLife Stadium & Estadio Azteca + sections.
`venue_metrics`	Per-section live telemetry (occupancy, ingress/egress, gate wait). Realtime-enabled.
`incidents`	Staff reports. AI classification stored as JSONB. Realtime-enabled.
`sops`	Seeded SOPs (medical, crowd, lost child, accessibility, weather, security).
`broadcasts`	Multilingual PA announcements composed by ops.
`chat_threads`	Reserved for persisted conversations (not required for the current session-based demo).

Real-time updates

venue_metrics and incidents are added to supabase_realtime. Both Fan and Ops pages subscribe to their venue channel and re-run TanStack Query fetchers when changes arrive.

Matchday simulator

POST /api/public/tick jitters every section's metrics ±5% and adjusts gate-wait ±30s. The Ops "Simulate matchday tick" button drives it live for demos.

3. Google Services used

Service	Where
Google Gemini (`google/gemini-3-flash-preview`) via Lovable AI Gateway	All chat, classification, ops brief, translation.
Google OAuth (managed by Lovable Cloud)	Sign in with Google on `/auth`.
Google Maps (deep link)	Fan "Open transit directions" link with lat/lng of venue. Google Maps Platform connector can be plugged in for embedded maps and Places API (New) without code changes; the app is designed to consume `VITE_LOVABLE_CONNECTOR_GOOGLE_MAPS_BROWSER_KEY` when connected.
Google Calendar (link template)	Fan "Add match to Google Calendar" — no OAuth required, uses public `calendar.google.com/render?action=TEMPLATE` URL.

4. Rubric mapping (Problem Statement Alignment)

Focus area	Where in the app
Code Quality	Strict TS, TanStack Start server-fn boundaries, Zod validation, `useSuspenseQuery` / TanStack Query, split component files (`StadiumMap`), typed AI SDK usage.
Security	Roles in separate table + `has_role()` security-definer, RLS on every table, `LOVABLE_API_KEY` server-only, Zod validation, HMAC-eligible cron route, `requireSupabaseAuth` middleware on every write server function, no `dangerouslySetInnerHTML`.
Efficiency	Gemini Flash (cheap + fast), Supabase Realtime instead of polling, TanStack Query cache, SVG map (no heavy 3D), lazy per-role routes, single provider per request.
Testing	Manual E2E is documented in Demo Script below; run `bun run build` and `bun run lint` in CI. Playwright hooks in place.
Accessibility	WCAG 2.2 AA color contrast, keyboard focus rings, skip-to-content link, ARIA live regions on chat + Ops brief, semantic landmarks, alt/`<title>`/`<desc>` on the SVG map, RTL support for Arabic, large-text toggle, `prefers-reduced-motion` respected.
Google Services	Gemini (chat/classification/brief/translate), Google OAuth, Google Maps deep link, Google Calendar link template.
Problem Statement Alignment	Every rubric track is covered — see the table in this README under "Rubric mapping".

5. Demo script (5 minutes)

/ — Landing page pitches the three roles.
/auth — Create an account (email or Google). You are now signed in as a Fan.
/fan — Pick venue + section. Change language to Spanish. Click "Nearest restroom → Get route" — the assistant replies in Spanish with a section-aware route. Toggle Large text.

Elevate yourself: in the Cloud dashboard (Users), add volunteer and ops roles for your user via SQL:

INSERT INTO public.user_roles(user_id, role) VALUES ('YOUR_UUID', 'volunteer'), ('YOUR_UUID', 'ops');

/staff — Log an incident: "Fan fainted in section 112, near aisle 4." Gemini classifies it as medical/high in ~1s. Move it across the board.
/ops — Click Simulate matchday tick twice. Watch the heatmap change. Click Generate on the AI Ops Brief — Gemini returns summary + risks + prioritized recommendations grounded in the metrics. Compose a broadcast in English, target ES/FR/PT, tone urgent, click Translate.

6. Assumptions

Telemetry is simulated via /api/public/tick — production would connect to real IoT/CV feeds.
fan role is auto-granted on sign-up; volunteer and ops are granted by an admin (via SQL / the Cloud dashboard) — no self-serve role escalation, by design.
Google Maps Platform connector is optional — the app links out to Google Maps rather than embedding, so the demo works without paid API keys.
No FIFA/tournament trademarks are used; visual language references the tournament thematically only.

7. Running locally

The project is a TanStack Start v1 app.

bun install
bun run dev       # starts Vite dev server on :8080
bun run build     # production build
bun run lint      # eslint

Environment (auto-managed by Lovable Cloud):

LOVABLE_API_KEY — for Gemini via Lovable AI Gateway (server-only)
SUPABASE_URL, SUPABASE_PUBLISHABLE_KEY, SUPABASE_SERVICE_ROLE_KEY — database + auth
Frontend uses VITE_SUPABASE_URL and VITE_SUPABASE_PUBLISHABLE_KEY

8. Tech stack

Framework: TanStack Start v1 (React 19 + Vite 7)
Styling: Tailwind CSS v4 + shadcn/ui + oklch design tokens
State: TanStack Query + Supabase Realtime
AI: Vercel AI SDK (ai, @ai-sdk/react, @ai-sdk/openai-compatible) → Lovable AI Gateway → google/gemini-3-flash-preview
Backend: Lovable Cloud (Postgres + Auth + RLS + Realtime), server functions via createServerFn, public webhook route via TanStack server routes
Validation: Zod
Markdown rendering: react-markdown

License

Prototype built for Google Prompt Wars Challenge 04. FIFA and tournament marks are property of their respective owners and are not used in this project.

👨‍💻 Developer

Saurabh Kumar

🔗 Connect With Me

⭐ If you found this project useful, consider giving it a Star!

Built with ❤️ using Google Gemini, React, TanStack Start and Lovable Cloud.

Name		Name	Last commit message	Last commit date
Latest commit History 2 Commits
.github/workflows		.github/workflows
.lovable		.lovable
public		public
src		src
supabase		supabase
tests		tests
.gitignore		.gitignore
.prettierignore		.prettierignore
.prettierrc		.prettierrc
README.md		README.md
bun.lock		bun.lock
bunfig.toml		bunfig.toml
components.json		components.json
eslint.config.js		eslint.config.js
package-lock.json		package-lock.json
package.json		package.json
tsconfig.json		tsconfig.json
vite.config.ts		vite.config.ts
vitest.config.ts		vitest.config.ts

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

PitchOps — Smart Stadiums & Tournament Operations (FIFA World Cup 2026)

⚽ PitchOps — Smart Stadiums & Tournament Operations (FIFA World Cup 2026)

🚀 Google Prompt Wars • Challenge 04

🌍 About The Project

🏆 PromptWars Virtual Participation

Build • Deploy • Win From Anywhere

📅 How The 14-Day Cycle Works

1. Approach & Logic

Why "one app, three roles"?

GenAI logic

2. How the solution works

End-to-end matchday flow

Data model

Real-time updates

Matchday simulator

3. Google Services used

4. Rubric mapping (Problem Statement Alignment)

5. Demo script (5 minutes)

6. Assumptions

7. Running locally

8. Tech stack

License

👨‍💻 Developer

© Made with ❤️ by Saurabh Kumar. All Rights Reserved 2025

Saurabh Kumar

🔗 Connect With Me

About

Uh oh!

Releases

Packages

Uh oh!

Uh oh!

Contributors

Uh oh!

Languages

Folders and files

Latest commit

History

Repository files navigation

PitchOps — Smart Stadiums & Tournament Operations (FIFA World Cup 2026)

⚽ PitchOps — Smart Stadiums & Tournament Operations (FIFA World Cup 2026)

🚀 Google Prompt Wars • Challenge 04

🌍 About The Project

🏆 PromptWars Virtual Participation

Build • Deploy • Win From Anywhere

📅 How The 14-Day Cycle Works

1. Approach & Logic

Why "one app, three roles"?

GenAI logic

2. How the solution works

End-to-end matchday flow

Data model

Real-time updates

Matchday simulator

3. Google Services used

4. Rubric mapping (Problem Statement Alignment)

5. Demo script (5 minutes)

6. Assumptions

7. Running locally

8. Tech stack

License

👨‍💻 Developer

© Made with ❤️ by Saurabh Kumar. All Rights Reserved 2025

Saurabh Kumar

🔗 Connect With Me

About

Topics

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Uh oh!

Contributors

Uh oh!

Languages

Packages