Civic AI: crowd-sourced urban digital twins

This work is coauthored.

Sense–Engage–Govern

A design framework for gamified, LLM-assisted crowd-sourced urban digital twins. It has three tiers and six dimensions, each a spectrum with a named tension, plus a privacy boundary. It draws on value-sensitive design, contextual integrity, and self-determination theory.

• contribution mode
• twin coupling
• incentive substrate
• verification locus
• temporal regime
• access threshold

Plus a privacy boundary

CS-UDT: a flood prototype for Jacksonville

Residents take part from a zero-install browser client, so they need no app. An LLM pipeline parses citizen reports and social posts into hazard events, and a convergence rule decides what counts as confirmed:

1. Three channels

   • Official sensor anomaly
   • Citizen report
   • Social post

2. Matched by place

   • The same map cell

3. What happens

   • Two channels agreenearby residents get a one-tap verification quest
   • All three agreethe event is confirmed

First three-channel confirmation: 10.8 h

1. 0 h
2. 6 h
3. 12 h
4. 18 h
5. 24 h
6. 30 h

Replay length: 30 hours.

At the storm’s peak: four confirmed cells and three open quests. The replay is simulated and all community content is synthetic.

Open the CS-UDT demo (simulated data).

Where it appears

• Oral presentation: DTRUE 2026, UF Jacksonville campus, Nov 19–20, 2026 (coauthored extended abstract).
• Full paper: EPiC Series in Technology (to appear).