Air quality systems today focus on isolated measurements—PM levels, VOC concentrations, or temperature thresholds—but real environments are far more complex. The same sensor readings can mean very different things depending on context: cooking, traffic pollution, human occupancy, or early combustion can produce overlapping signals that traditional rule-based systems fail to interpret correctly.

I decided to build EnviroFusion-Q to move beyond raw air quality numbers and instead teach a device to understand what kind of environment it is experiencing. The motivation behind this project is simple: meaningful environmental intelligence only emerges when multiple heterogeneous sensors are fused and interpreted together, directly at the edge.

EnviroFusion-Q works by combining gas and VOC data from the ENS160, particulate matter dynamics from the GP2Y1014AU0F dust sensor, and thermal-humidity context from the AHT21. These signals are sampled synchronously and analyzed over short time windows. Rather than evaluating each sensor independently, the system extracts temporal patterns and cross-sensor relationships that form a unique “environmental fingerprint.”

Using Edge Impulse, these fused features are used to train a compact neural network that runs entirely on the Arduino Uno Q. The model classifies complex real-world environments—such as clean indoor spaces, cooking activity, traffic pollution, smoke events, or abnormal combustion—in real time, without relying on cloud connectivity.

The result is a fully self-contained edge AI system that transforms raw sensor data into actionable environmental context, demonstrating how true sensor fusion enables smarter, more reliable decisions on microcontroller-class hardware.