Edge AI for wildlife monitoring that learns without the cloud

Biodiversity monitoring typically requires resource-intensive manual surveys or cloud-dependent AI systems unsuitable for remote field deployments with limited power and connectivity. This work proposes knowledge adaptation—replacing implicit model parameters with an explicit knowledge base that separates visual encoding from reasoning. The architecture allows field devices to improve performance by updating structured knowledge rather than retraining models, enabling sustainable on-device inference. Developed through collaboration with biologists and Indigenous communities, the system demonstrates how edge AI can support ecological monitoring while respecting local expertise and reducing infrastructure demands.