A procession through the heart of Washington, DC — run from a phone.

The challenge. A major annual procession moves through central Washington on closed streets with a permit, a hard date, and roughly sixty rostered volunteers spread along a formation that is itself moving. Coordination ran on group chats, a link-list page, and a shared spreadsheet of phone numbers.

Why the usual answers failed. No shared picture of where the procession front was. No way to call for help discreetly versus loudly. No system for gear that goes out and must come back, for rides to the assembly point, or for verifying fundraiser payments sent as screenshots.

The stakes. Thousands of phones on congested cell towers, volunteers of every age and skill, a safety-adjacent public event in a capital city — and a date that doesn’t move.

What we built. One installable app with four faces. The public sees a countdown, a live map with the moving procession front, parking help, and a way to ask a question. Volunteers check in, raise a discreet or loud alert with one hold, talk over push-to-talk radio, and sign out gear. Leads triage alerts with a siren takeover, see their roster, and plan relief. Admins edit everything — branding, route, teams, inventory — as live configuration.

The workflow. An onlooker scans a QR and sees the live map. A volunteer taps once to raise help. A lead acknowledges on a phone, sees who’s nearest, and speaks to a team. A supporter sends a payment screenshot; the AI extracts the fields, deterministic checks verify amount, recipient, and memo, and the order reconciles itself.

Decisions that mattered.

• Offline-first, maps cached. Thousands of phones share one tower; the app assumes it.
• Silent versus loud. “Send help quietly” never alarms the crowd.
• The AI is an extractor, not a judge. Payment proofs are trusted by hard checks, not by the model.
• Privacy-first analytics. No IP addresses stored; the organization learns what it needs and nothing more.

Timeline. First commit to public launch in six weeks — including a five-role end-to-end certification campaign and a community promo film.

Outcomes. The procession ran on it. Launch-window analytics from production: 792 views, 141 unique visitors, 47 installed-app sessions. Forty-six real orders with thirteen AI-verified payment proofs. The team reports that users and volunteers praised the app; it remains live. After the event, the team asked for three features — all shipped within two days.

What this proves. A white-label event-operations engine: a new organization or event is a configuration file. The carpool module, radio stack, and payment verifier each stand on their own.