Water Leakage Prediction & Visualization

• 

  Water Leak

Inspiration

Water scarcity is becoming one of the world’s most pressing challenges by 2030, global demand for freshwater is projected to exceed supply by 40%. Cities lose up to 30% of their treated water due to undetected leaks in underground networks. We were inspired to build a system that not only detects these leaks early but also helps municipalities visualize and act on real-time insights. Our goal: contribute to sustainability through technology and data-driven decision-making.

What it does

Water Leakage Prediction & Visualization is an AI-powered platform that predicts water leaks in city pipelines and displays them on an interactive Google Maps dashboard. It: Uses Vertex AI and Gemini for intelligent predictions Is deployed on Cloud Run for scalability and real-time inference Visualizes leaks and KPIs such as water saved, cost reduction, and emission impact Empowers city managers with instant, data-backed alerts

How we built it

Data Collection: We used open datasets (Kaggle, UCI) and simulated pressure–flow readings to create leak scenarios. Preprocessing: Data was cleaned and structured using BigQuery for integration with Vertex AI. Model Training: Vertex AI AutoML & custom XGBoost/RandomForest models were trained to predict leak probability. Deployment: The model was deployed on Cloud Run via a FastAPI backend connected to the Vertex AI endpoint. Frontend: Built with React.js and Google Maps API to visualize leaks and live updates from Firebase Firestore. Integration: Real-time simulated data flows → Vertex AI → Cloud Function → Firestore → React Dashboard.

Challenges we ran into

Data limitations: Real-world water network datasets are rare, so we simulated realistic sensor data. Model latency: Ensuring low-latency predictions from Vertex AI when deployed via Cloud Run. Maps integration: Rendering live leak updates efficiently without overloading the browser. Time pressure: Building a full ML-to-frontend pipeline within hackathon time constraints.

Accomplishments that we're proud of

Developed a fully functional demo integrating AI + Google Maps visualization. Achieved 90%+ prediction accuracy on simulated data. Successfully deployed the ML model on Google Cloud Run for real-time API inference. Created a clear and impactful dashboard showing water savings and leak locations. Delivered a project that aligns with UN SDG 6 (Clean Water & Sanitation).

What we learned

How to integrate Vertex AI, BigQuery, and Cloud Run for end-to-end ML pipelines. How to connect AI outputs with live visual dashboards in React + Firebase. The importance of data realism and validation when simulating IoT environments. The value of team collaboration and agile task division under tight deadlines.

What's next for Water Leakage Prediction & Visualization

Integrate Gemini for natural language reporting, e.g., “3 leaks detected today, 1200 liters saved.” Collect real IoT sensor data from water utility partners. Enhance the model using deep learning (LSTM) for time-series leak prediction. Launch a city-scale pilot project to support smart city water management systems. Publish open-source dashboards to encourage civic innovation in sustainability tech.

Built With

• ai
• api
• bigquery
• cloud
• css
• fastapi
• firebase
• firestore
• gemini
• google
• html
• javascript
• maps
• python
• randomforest
• react.js
• run
• storage
• vertex
• xgboost