Skip to content

Engineer focused on aerodynamics, composite structures, and applied ML. I like hard problems. The ones where the physics fights back.

Early Bushfire Detection on IoT Sensors (GGM-VAE & Hybrid-VAE)

For my Honours Project, two unsupervised GRU-based VAEs on eCO₂ and TVOC. A compact GGM-VAE and a Hybrid STFT-CNN-GRU-VAE both detect all ignition events in about four minutes on average; the GGM-VAE wins on F1 and false positives and is deployed on an STM32WLE5 LoRa node.

Algorithm pipeline
VAE loss curves
Microcontroller and sensors

↳ Experiment-level hit rate 100 percent across eight ignitions with average first detection about 4 minutes

ML · Anomaly Detection · IoT · Time-series
View project