AquaLERT

• 

  AquaLERT

Inspiration: The Silent Crisis

Our inspiration for AquaLERT comes from a critical, life-threatening challenge facing Haiti: the water crisis. Over 4 million people lack access to a basic water source, and preventable waterborne diseases like cholera and dysentery remain a constant threat. We realized the core problem wasn't just a lack of water, but a critical lack of information. Traditional lab testing is slow, prohibitively expensive, and inaccessible to the very communities that need it most. How can you protect yourself if you don't know the water you rely on is unsafe? We were inspired to build a solution that bridges this information gap, using modern AI to put the power of a water testing lab directly into the hands of the people.

What It Does: Our Three-Pillar Solution

AquaLERT is a complete, end-to-end water intelligence platform designed to provide immediate, actionable insights. Our solution is built on three core pillars:

1. Instant Field Testing

A user in the field can input readings from a low-cost sensor kit into our web app. Our highly-tuned LightGBM machine learning model analyzes nine critical parameters and delivers a clear "Potable" or "Not Potable" prediction in under 30 seconds.

2. AI-Powered Advisory

A simple prediction isn't enough. We integrated Google's Gemini API to translate the complex scientific data into a detailed, easy-to-understand public health advisory. It explains the risks, provides recommended actions for treatment and control, and, most importantly, delivers this guidance in both English and Haitian Creole to ensure it is accessible to everyone.

3. Community Monitoring & Hotspot Detection

Every test submitted is automatically geotagged and added to our Live Interactive Map. This creates a powerful, crowdsourced intelligence network. NGOs and public health officials can use our dashboards to see contamination hotspots in real-time, track trends, and deploy resources with data-driven precision. We also included an innovative Visual Analysis tool that allows users without sensors to get a preliminary safety assessment by simply uploading a photo of their water.

How We Built It

AquaLERT is a full-stack application built with a focus on robustness, scalability, and user experience.

Frontend: We used Streamlit for its ability to rapidly create beautiful, interactive, and data-centric user interfaces. We integrated Folium for the live mapping component and Plotly for our dynamic data visualizations and dashboards.

Backend: A lightweight Flask server acts as the heart of our application. It exposes a simple API that handles requests from the frontend, processes data, and interacts with our AI models.

Machine Learning & AI:

Classifier: We trained and fine-tuned a LightGBM model on a water potability dataset. We went through a rigorous process, starting with baseline models and improving performance by using robust preprocessing pipelines (SimpleImputer with a median strategy) and systematic hyperparameter tuning (RandomizedSearchCV) to achieve high accuracy.

Generative AI: We used the Google Gemini API for our advisory feature. We engineered specific prompts that take the ML model's output and the raw data as context to generate safe, relevant, and multilingual health recommendations.

Forecasting (Proof-of-Concept): We built a strategic forecasting module using Facebook's Prophet to demonstrate how AquaLERT could predict future water degradation, a key feature for long-term planning.

Challenges We Ran Into

Our biggest challenge was moving beyond a basic, low-accuracy machine learning model. Our initial models were only slightly better than a coin flip. We overcame this by systematically diagnosing the problem (class imbalance and naive data imputation) and engineering a more robust pipeline. Experimenting with different techniques like SMOTE and class weighting, and ultimately using RandomizedSearchCV with LightGBM, was key to building a classifier we could trust.

Another challenge was designing a user interface that could present complex scientific data in a way that was instantly understandable to a non-technical user, which led to the creation of the AI Advisory and the clear, color-coded results.

What's Next for AquaLERT

AquaLERT is more than a hackathon project; it's a scalable platform for social good. Our next steps are: Hardware Integration: Develop and document a low-cost, solar-powered sensor kit that integrates directly with our platform.

Deploy Forecasting Module: Fully integrate the strategic forecasting model to provide long-term risk analysis to our partners.

NGO Partnerships: Collaborate with on-the-ground organizations in Haiti to pilot AquaLERT in the field and gather real-world data.

Offline First: Enhance the application to have full offline capabilities for data collection in areas with no internet access.

Built With

• css3
• flask
• folium
• google-gemini-api
• html5
• lightgbm
• pandas
• plotly
• prophet
• python
• scikit-learn
• streamlit