EdgeSense

Cloud-free CO₂ and PM2.5 forecasting running entirely on an ESP32-S3, using tree-ensemble models compiled to pure C for real-time inference.

EdgeSense is a cloud-free indoor air quality forecasting system that performs real-time prediction of CO₂ and PM2.5 directly on an ESP32-S3 microcontroller, eliminating the need for internet connectivity or cloud inference. The system uses embedded ensemble tree models — Random Forest, XGBoost, and LightGBM — trained offline and converted into pure C for deployment within strict memory and latency constraints.

A unified on-device feature pipeline combines temporal encodings, lagged sensor values, and rolling aggregates to enable accurate short-term forecasting. Experimental evaluation shows that boosted models achieve higher accuracy than Random Forest while remaining resource-efficient, demonstrating that reliable IAQ prediction is feasible on ultra-low-cost hardware.

The project highlights a complete edge-AI pipeline that prioritizes privacy, deterministic performance, and scalability for classrooms and resource-constrained indoor environments.