city-simulator

A Phaser-based isometric city simulator that A/B-compares routing strategies for restaurant choice, with a real-time Google Maps heatmap of crowd density and six comparison charts of user-experience metrics. Built as the visualization piece of the Crowdless project.

🥇 Part of the team's 1st place entry at AI Hackathon 2026 — Open Track (ACE @ AUEB, March 2026).

What it is

Two sims run side by side, deterministically (seed crowdless2026):

Side	Strategy	Picks restaurants by...
Baseline — No App	NearestRestaurantGoal	Closest physical restaurant only
Crowdless AI — With App	AIRecsRestaurantGoal	Distance + current crowdedness + queues

NPCs spawn, pathfind through the isometric city, queue at restaurants, and either get served or abandon. A live Google Maps overlay heatmaps crowd density per restaurant in real time. Six Chart.js charts compare the two sims on:

• Walk distance to store
• Crowdedness at arrival
• Queue length at arrival
• Queue abandonments
• Customers served per store
• Bad-experience events per user

The visualization made the case for what a crowd-aware app like Crowdless does to a city's flow.

⚠ Honest framing — the "AI" is not ML

AIRecsRestaurantGoal.js is a deterministic, hand-coded heuristic that emulates what a Crowdless recommender would output. There is no trained model, no learning loop, no training data. It's a stand-in designed to faithfully represent the intent of the app being demoed.

We built this as the visual WOW factor for the hackathon pitch. It's a real simulator — pathfinding, queues, abandonment, metrics, all implemented — but the "AI" lever is a hand-tuned policy, not a learned one. Treat it accordingly.

Stack

• Game engine: Phaser 3 (arcade physics build, loaded via CDN)
• Maps + heatmap: Google Maps JavaScript API with visualization and geometry libraries
• Charts: Chart.js (CDN)
• Code: Plain ES modules — no bundler, no package.json
• Map data: map.json (Tiled-format, ~280 KB), centered on Athens (37.9838, 23.7275)

Run locally

You'll need a Google Maps JavaScript API key with visualization and geometry libraries enabled. Restrict the key to your dev referrers (http://localhost, http://127.0.0.1) before using it anywhere — never ship an unrestricted Maps key.

1. Open index.html and replace the placeholder:

   <script async
     src="https://maps.googleapis.com/maps/api/js?key=YOUR_GOOGLE_MAPS_API_KEY&libraries=visualization,geometry">
   </script>

   with your own key.
2. Serve the directory over HTTP (Maps + module imports won't work from file://):

   python -m http.server 8000
   # then open http://localhost:8000

Team

Same five-person team as Crowdless:

• Alexandros Gkiafis (@AlexTuring010) — simulator (this repo)
• Theodoros Exarchos (@EXARTeo)
• Mariglena Reci
• Stelios Rotas
• Panagiotis Thivaios

Links

• Main app repo (Cassini-era): https://github.com/AlexTuring010/crowdless
• AI-Hackathon-winning rebuild (private): https://github.com/EXARTeo/Crowdless_App
• Crowdless project page: https://www.linkedin.com/company/crowdlessapp/
• AI Hackathon 2026 winners post: https://www.linkedin.com/posts/aihackathon2026-opentrack-innovationecosystem-ugcPost-7436487613802381312
• Prior year of the same hackathon series (different team, different challenge, didn't win): https://github.com/AlexTuring010/novibet-clone

License

MIT — see LICENSE.