Landslides pose great risks to human beings and infrastructure, thus requiring powerful real-time monitoring systems. This paper suggests a TinyML architecture on the edge using a cluster-style wireless sensor network (WSN) with low-cost ESP32 microcontrollers. Integrating soil moisture, tilt, vibration, temperature and humidity sensors, energy-efficient methods have been applied such as adaptive power control and duty cycling, to maximize the working time. An edge-optimized 1D CNN model is a lightweight solution that can classify risk in real-time on the edge. Besides, nodes automatically form clusters based on the RSSI and battery conditions to provide fault tolerance and fair energy consumption. Intra-cluster communication depends on ESP-NOW and neighbor table update and Cluster Head (CH) re-election mechanisms are used to ensure reliability in the network. The empirical justification in efficient classification of landslide risk levels as Low, Medium, and High with limited resource consumption and latency time is depicted. Future and continued directions of research involve exploration of complex machine-learning techniques and inter-cluster communication protocols.