Wildlife has been in danger for quite some time from various kinds of human activities, ranging from directly destroying habitat to impact from climate change caused by human's unsustainable way of living. Many organisations have established conservation projects and activities to try to mitigate the effects and save wildlife, one of the most important activities of which is monitoring species. Automatic sound recognition systems have proven to be an effective tool used during conservation activities. There are several devices out on the market which can be installed in the wild and record animal sounds. However, they are inaccessible due to being expensive and they only record sounds and require further proprietary software to classify sounds elsewhere. The intention of the project is to create an alternative low-cost device that can be installed in the wild and be able to both record and classify animal sounds right on the edge. This involved developing a machine learning model from scratch that is able to classify bird sounds. The project also created edge frameworks for two architectures, Raspberry Pi and Arduino to which the machine learning model is deployed to. As well, 3D case solutions were designed allowing for both architectures to be safely deployed in the wild. The project then created a gateway device and a framework for it which is used to store results transmitted by edge devices. Afterwards, the created model was evaluated against competitor's model which showed to have a competitive performance, outperforming competition in some cases. The project has also investigated performance of the model running on both architectures and compared architectures to understand which one is more suitable to use when. To achieve the outlined goals, the project has tackled the development of loT applications, fundamentals of machine learning, architectural design and development of computer aided designs.