
CityCascade AI is an interactive disaster cascade simulator a city breaking apart in real-time, and an AI healing it. THE PROBLEM When Dhaka floods, no single authority sees the full picture. The weather office sees rain. Hospitals see patients. Traffic sees jammed roads. Nobody sees the chain: flood, blocked ambulances, contaminated water, dengue outbreak two weeks later. This siloed blindness costs lives. WHAT WE BUILT Select a disaster (flood, fire, heatwave), click any of 19 real Dhaka neighborhoods, and watch a BFS cascade algorithm propagate the crisis across the city's adjacency graph, color-coded by severity in real-time. Each zone carries real data: elevation, drainage quality, population density, hospital load, and disease risk multipliers. The engine computes flood depth, hospital capacity loss, road blockage, and contamination risk per zone per hop. One click sends the full cascade state to Gemma AI, which generates a specific response plan with actual zone names, road IDs, hospital overflow allocation, ambulance reroutes, and water boil advisories. Not generic advice. THREE MISSIONS, ONE CHAIN EarthCare: Flood simulation, water contamination, drainage failure modeling. HealthNet: Hospital surge analysis, disease risk prediction, emergency routing. SmartCity: Road disruption visualization, safe route recalculation, crisis dashboard. Not separate tabs. One causal chain: environmental crisis, health impact, city response. GEMMA 3 AI providers: Gemini 1.5 Flash, OpenRouter Gemma 3 4B free tier, and Ollama Gemma 2 for fully offline inference. The AI receives structured cascade state, not a vague prompt, producing plans judges can verify against the live map. STACK Next.js 16, React 19, TypeScript, Leaflet.js, Tailwind v4, MongoDB Atlas, BFS graph engine, Gemma via OpenRouter and Ollama, Judges draw a flood. Mirpur-10 turns red, Pallabi goes orange, the road to Dhaka Medical greys out. Gemma tells them exactly what to do. That's the moment nobody forgets.
10 May 2026