ecogrid

Inspiration

Climate data is fundamentally fragmented. Emergency planners and decision-makers are constantly drowning in isolated weather reports, struggling to tie real-time meteorological telemetry directly to actionable interventions. We wanted to build something that bridges this gap, moving away from static charts into an interactive, planetary-scale command center that unifies live global data streams, 3D spatial rendering, and in-browser machine learning.

What it does

EcoGrid AI tracks real-world heat stress and water deficit anomalies pulled directly from live telemetry feeds. Users can inspect high-risk sentinel regions, review baseline vulnerability metrics, and deploy targeted interventions through a tactical command matrix. The system features a fully predictive, unrigged autonomous AI agent that evaluates risk horizons dynamically, balancing global resources and triage to keep planetary integrity stable without artificial cheats.

How we built it

• Frontend & Architecture: Built with TanStack Start, Tailwind CSS, and React Query for a high-performance, reactive user experience.
• 3D Globe Rendering: Designed a custom Three.js vector globe featuring real-time strike vectors and spatial coordinate rendering to monitor global climate nodes.
• Telemetry & ML: Integrated live Open-Meteo telemetry streams and engineered a custom in-browser L 2 ridge regression model that trains directly inside the browser environment on every simulation tick: $$\min_{\beta} |y - X\beta|_2^2 + \lambda|\beta|_2^2$$
• Autonomous Decision Engine: Developed a multi-factor predictive agent utilizing rolling horizon forecasts, cost-benefit resource scarcity balancing, and ripple-stress contagion modeling.

Challenges we ran into

• Real-Time Data Sync: Managing heavy live meteorological telemetry streams without degrading browser rendering performance required optimizing our Three.js render loops and state management.
• Autonomous Economic Balancing: Tuning the simulation loop so that resource regeneration and the autonomous AI agent didn't fall into death spirals during multi-region cascading failures, ensuring the agent makes mathematically sound, unrigged triage decisions.

Accomplishments that we're proud of

• Successfully integrating real-time Open-Meteo weather data directly into a custom 3D Three.js interactive globe entirely inside a browser environment.
• Engineering a fully autonomous, predictive decision-making agent that runs complex multi-step lookaheads and resource optimization locally on every tick without UI lag.

What we learned

We mastered bridging the gap between chaotic global climate telemetry and clean, human-centered UI design. Building custom 3D spatial environments alongside real-time client-side machine learning and complex state dynamics taught us how to engineer high-performance, transparent systems entirely inside the browser.

What's next for ecogrid

• Expanding our predictive models to factor in secondary environmental indicators like soil moisture and urban heat island effects.
• Adding multi-region comparative forecasting to help policy makers simulate cross-border mitigation strategies.

Built With

• reactquery
• tailwand
• three.js