3D City simulation
Landing Page
Loading page
Conversations on the street throughout the city
Building damage to the city after a fire

Inspiration

We all wanted a project that improves societal behavior. We wanted to create a website that researchers and analysts could use for a wide range of possible life-changing events.

What it does

Users can make a custom 3D map with live AI agents that experience user-inputted events. These events can be as small as roadside festivals, to policy changes, to alien invasions. After the event occurs, an overview is shown with how the moods of the agents were affected, and what they learned from the event. The user can then decide to continue from that point after the event, or return to a past timeline before the event.

All simulations overviews can be viewed in the database at a later time, and researches can use this information to improve their towns. Surveys and data analysis can be difficult to collect, but this makes it simpler and more accessible to simulate events.

How we built it

Frontend: Claude Opus 5.5, json, html, ts, Google, ChatGpt - Background images were found on Google and recreated using ChatGpt. We wrote out 13 pages (nearly 18k words) of prompting given to Claude in Cursor. Backend: python, ts, css, javascript, codex, ChatGPT 5&6, Claude Opus 5.5, Grok 4.6, Gemini API, ElevenLavs, TigerData, Snowflake.

Challenges we ran into

Creating an aesthetically pleasing 3D-interactive map with 3D agents Inputting and interpreting custom events Zooming and listening in on conversations

Accomplishments that we're proud of

Small details in UI to allow users to have an easy experience, along with the visual aesthetics that are coherent throughout the website The visuals of live conversations between agents around the map, and being able to zoom in on them.

What we learned

Time management, debugging, collaboration, task delegation

What's next for Agent City

Some more debugging with ElevenLabs, as it's running slowly when trying to listen in on conversations between agents. Setting up specific user accounts. Increasing realness in conversations and event outcomes. Enhancing 3D visuals of the map and agents. Ability to make custom improvements to the town, and notice the change in outcome for similar events that occurred before the improvements.