News

Scalable AI-Driven Monitoring Data Pipeline 

“Design and Evaluation of a Scalable Data Pipeline for AI-Driven Air Quality Monitoring in Low-Resource Settings” - Cornell University

• Presents an open-source cloud-native ETL system that ingests data from low-cost sensors, weather APIs, and reference monitors. 

• Uses AI calibration and forecasting for real-time analytics with millions of incoming measurements.

• Includes practical deployment insights and is particularly relevant for urban air quality platforms in developing regions.

Embedded Machine Learning Systems for Monitoring

 “An embedded machine learning system method for air pollution monitoring and control” - AIMS Press

•  Proposes an embedded ML approach (ORCS-ASVM) to improve accuracy of pollutant detection (PM₂.₅, NO₂, SO₂). 

• Good real-time performance and lower computing cost compared to traditional models.

Veli: Unsupervised Method for Low-Cost Sensor Correction (2025)  - Cornell University

• Introduces Veli, an unsupervised Bayesian correction model that improves the accuracy of low-cost air quality sensor readings without co-location with reference stations.

• Separates true pollutant signal from noise and drift and introduces the first large benchmark dataset (23,737 sensors). 

• Helps scale dense monitoring networks with real-world robustness. 

30-Month Field Evaluation of Low-Cost CO₂ Sensors (2025) 

-European Geosciences Union

• Long-term field calibration of low-cost NDIR CO₂ sensors against high-precision instruments in a dense urban network. 

• Shows how environmental correction and drift calibration can yield reliable CO₂ monitoring at ~10x lower cost.