Ecovision_ChatGPT did this

Inspiration

Buildings such as shopping malls consume large amounts of electricity, water, and cooling energy every day. However, facility managers often struggle to understand where energy is being wasted because the data is usually scattered across different systems and shown only as raw numbers. We were inspired to build Ecovision to make building sustainability data easier to understand, visualize, and act on. Our goal was to help businesses identify abnormal energy usage, reduce operational waste, and support ESG goals through a more intelligent and interactive dashboard.

What it does

Ecovision is a sustainable building intelligence platform that helps users monitor energy, CO₂ emissions, water usage, waste, and building performance. The system includes a dashboard with sustainability graphs, a 3D building heatmap, and an ML-based anomaly detection module. Users can view building data across different floors and rooms, compare usage patterns over time, and identify areas with unusual energy consumption. The ML anomaly detection model predicts the expected energy usage for a specific floor, room, and hour. It then compares the expected value with the actual energy usage to detect anomalies. Instead of only showing that something is wrong, Ecovision also explains the possible reason, severity level, recommended action, business benefit, and ESG alignment.

How we built it

We built the frontend using React, Vite, Recharts, Three.js, and @react-three/fiber. The dashboard displays analytics graphs, sustainability metrics, and an interactive 3D building heatmap to help users visually understand resource consumption across the building.

For the machine learning module, we created a separate backend system using Python, FastAPI, and scikit-learn. We trained a Random Forest Regressor to predict expected hourly energy usage based on features such as floor, room, time, occupancy, operating hours, visitor count, weather conditions, and tariff rate.

The model uses the difference between actual and expected energy usage to calculate an anomaly score and severity level. We also added explainable output logic so the system can classify possible issues such as HVAC overcooling, lighting after operating hours, standby plug load, water spikes, or high total energy usage. Finally, we connected the ML backend to the React frontend so the anomaly results could be displayed directly inside the dashboard.

Challenges we ran into

We faced integration challenges when connecting the FastAPI backend with the React frontend. The frontend had to fetch ML results, display severity clearly, and show detailed explanations without recalculating the anomaly logic on the client side. We also had trouble with pushing our dataset files into GitHub because it is too large in size. Other than that, we also faced issues when pushing and merging our files together on GitHub, there were cases where it will run on one person's laptop but not another.

Accomplishments that we're proud of

We are proud that Ecovision is not just a static dashboard, but an interactive decision-support tool. It combines analytics, 3D visualization, and machine learning into one system. We are also proud of building an explainable anomaly detection module. Instead of only flagging abnormal usage, the system provides reasons and suggested actions that a facility manager can understand and use. Another accomplishment is successfully integrating the ML backend with the frontend dashboard. This made the prototype feel more complete because users can directly see real anomaly insights inside the application.

What we learned

We learned that sustainability problems are not only about collecting data, but also about presenting it in a way that helps people make better decisions. A dashboard becomes much more useful when it explains what is happening, why it may be happening, and what action should be taken. We also learned the importance of context-aware machine learning. Energy usage cannot be judged fairly using simple fixed thresholds because building usage changes depending on time, occupancy, weather, and room type. From a technical perspective, we learned how to combine frontend visualization, backend APIs, and machine learning into one integrated system.

What's next for Ecovision_ChatGPT did this

Next, we plan to improve Ecovision by connecting it to real-time IoT or building management system data so that anomalies can be detected live. We would also like to connect it to an actual backend database such as MySQL or MongoDB. We also want to expand the ML system with a second model that recommends optimization actions, such as adjusting cooling schedules, shifting non-critical loads, or identifying areas with the highest saving potential. In the future, Ecovision could include automated ESG reporting, carbon reduction tracking, alert notifications, and a chatbot-style assistant that helps facility managers understand the dashboard and take action faster.

Built With

• css
• fastapi
• fiber
• javascript
• jsx
• python
• randomforestregressor
• recharts
• three.js
• vite
• zustand