AirSense AI: Urban Air Quality Intelligence Platform
Abstract
Urban air pollution is one of the most persistent environmental and public-health problems facing Indian cities, and Delhi in particular records some of the highest particulate matter concentrations in the world during the post-monsoon and winter months. Public air-quality portals in India largely display current or historical readings, but they rarely tell a city administrator what the air will look like over the next few days, or where a pollution spike is likely originating from. This gap between “knowing today's AQI” and “deciding what to do about it” motivates AirSense AI, a full-stack urban air-quality intelligence platform designed as an academic prototype for Delhi. The system ingests PM2.5, PM10 and NO₂ station readings together with weather variables, satellite fire detections and land-use information, converts pollutant concentrations into a standardized Air Quality Index using the U.S. EPA methodology, and applies a gradient-boosted decision tree model (XGBoost) to forecast AQI up to 72 hours ahead. In parallel, the platform combines NASA FIRMS active-fire data with OpenStreetMap land-use polygons and Haversine-distance calculations to rank probable pollution hotspots, and uses a large language model (Groq/LLaMA 3) to translate the resulting numerical evidence into readable, bilingual (English/Hindi) advisories and enforcement summaries for administrators and citizens. The results are presented through a React and Leaflet-based interactive dashboard. This paper describes the motivation, related work, and proposed architecture of AirSense AI as an integrated decision-support system rather than a monitoring-only tool. As the platform is at the prototype stage, no experimental forecasting accuracy, source-attribution validation, or user-study results are reported here; these are identified as required additions for the next version of this paper.