Inspiration
Indoor air quality is one of the most overlooked factors affecting health in schools, colleges, offices, and public buildings. Poor ventilation and elevated CO₂ levels can increase discomfort and contribute to the spread of airborne diseases, yet most institutions lack affordable, intelligent monitoring systems.
We wanted to build a solution that goes beyond displaying sensor readings. Our goal was to create an AI-powered platform that helps administrators understand risks, evaluate interventions, and receive actionable recommendations before problems escalate.
AirGuardian AI was built to transform environmental data into meaningful decisions that support healthier indoor spaces.
What it does
AirGuardian AI is an intelligent Indoor Air Quality (IAQ) monitoring platform that combines machine learning and generative AI to help monitor classroom environments.
The platform provides:
🏫 Live monitoring across multiple classrooms 🤖 Machine learning–based infection risk classification 📈 Future air quality forecasting 🧠 Explainable AI using SHAP 🌐 Digital Twin simulation for "what-if" scenarios 🎯 Reinforcement Learning–based ventilation recommendations 🚨 Sensor anomaly detection 📊 Building Health Score for overall facility monitoring ✨ Gemini-powered executive health summaries 📩 Automated notifications using ViaSocket 📄 PDF report generation for administrators
Instead of presenting raw numbers, AirGuardian AI converts environmental data into actionable insights that support informed operational decisions.
How we built it
The project follows a modular AI pipeline:
Sensor data is processed and cleaned. A Machine Learning model predicts classroom infection risk. Building Health aggregates room-level predictions into a facility-wide score. Forecasting estimates future air quality trends. SHAP explains why predictions were made. A Digital Twin simulates environmental changes. A Reinforcement Learning agent recommends ventilation strategies. Gemini generates executive summaries from the AI outputs. ViaSocket distributes alerts automatically to stakeholders. Tech Stack Python Streamlit Scikit-learn Pandas NumPy Plotly SHAP Google Gemini API ViaSocket Joblib SQLite Challenges we ran into
Building multiple AI components into a single workflow was the biggest challenge.
Some of the challenges included:
Designing a modular architecture that kept each AI component independent. Combining predictive models with explainable AI. Integrating Gemini without allowing it to alter machine learning predictions. Creating automated alert workflows through ViaSocket. Maintaining fast dashboard performance while running multiple AI modules. Accomplishments that we're proud of Successfully integrated multiple AI techniques into one application. Built a complete end-to-end monitoring workflow. Added explainability instead of treating AI as a black box. Implemented automated alerting through ViaSocket. Created an intuitive dashboard suitable for non-technical users. What we learned
Throughout this project we gained experience with:
End-to-end machine learning deployment Explainable AI (SHAP) Reinforcement Learning Time-series forecasting Digital Twin concepts Generative AI integration Workflow automation using ViaSocket Building production-style Streamlit dashboards What's next for AirGuardian AI
Our future roadmap includes:
Real-time IoT sensor integration Multi-building monitoring Edge AI deployment Mobile application Predictive maintenance Smart HVAC integration Energy optimization Smart city deployment
Log in or sign up for Devpost to join the conversation.