CityPulse AI
Urban Defense Grid
An autonomous, multi-agent intelligence platform designed to protect urban populations from environmental health risks in real-time.
Terminal Activity
Booting Multi-Agent orchestrator...
Loading environmental prediction models...
Establishing CUDA GPU connections... SUCCESS
Ingestion Agent active: Monitoring AQI
Triage Agent active: Parsing citizen signals
DB Cache TTL Engine initialized (10 min ruleset).
Awaiting zone selection for live analysis...
What is this?
The Multi-Agent Urban Defense Grid
1. Robust Gemini Integration
The entire intelligence layer is powered by Google Gemini 2.0 Flash. Crucially, the AI integration utilizes a resilient lazy-loading architecture that prevents Next.js caching bugs, ensuring environment variables always resolve perfectly at runtime.
2. Agent Swarm via LangGraph
The Ingestion and Triage agents actively monitor city APIs. Forecast predicts the future, while Decision and Reflection synthesize mitigation plans in a strict LangGraph state machine.
3. HITL Architecture
The system features a strict Human-in-the-Loop (HITL) checkpoint. High-risk decisions are paused and sent to the dashboard's Approval Queue for human oversight. If rejected, a Learning Agent extracts new behavioral rules for future use!
4. GPU What-If Simulations
City officials can execute "What-If" simulations directly from Mission Control. Backed by cuDF and cuML on Nvidia hardware, traffic changes can be simulated in milliseconds, generating fully revised AI decisions instantly.
Why this?
The Urban Health Crisis
Cities are growing rapidly, and with them, environmental risks like severe air pollution, heatwaves, and localized chemical hazards. Traditional dashboards only show what has already happened.
City officials are overwhelmed by data. When AQI spikes to hazardous levels, they don't just need a red line on a chart—they need to know exactly which schools to close, which hospitals to alert, and how to route emergency services.
CityPulse AI was built to solve the paralysis of analysis. It doesn't just show data; it actively simulates the future and drafts the exact mitigation plans needed to save lives.
How we solve it
Deep Tech & GPU Acceleration
Nvidia CUDA Backend
While the agents make decisions, they rely on a high-performance Python FastAPI backend. This service utilizes GPU-accelerated Pandas (cuDF) to run massive Monte Carlo simulations and "What-If" scenarios in milliseconds.
200 OK - 45.2ms
GPU Memory: 4.2GB / 24GB
500,000 rows processed.
1. Real-time Ingestion
Connecting directly to IoT sensors and citizen reporting APIs.
2. Predictive Modeling
Forecasting AQI drift based on wind, traffic, and industrial output.
3. Autonomous Action
Drafting emergency orders, dispatching medical alerts, and rerouting traffic.
4. Smart Caching Layer
Dynamic 10-minute DB cache prevents redundant LangGraph triggers while maintaining strict real-time accuracy.