Access to safe drinking water remains a critical challenge in rural Nigerian communities, where contamination goes undetected due to lack of laboratory infrastructure and infrequent manual testing. Existing monitoring approaches rely on expensive equipment or constant internet connectivity - neither of which is realistic in low-resource field settings. This work presents an ongoing effort to develop a real-time, internet-free water quality monitoring system that is affordable, deployable, and AI-powered. The system is built around three low-cost sensors: pH, turbidity, and temperature, with a target hardware cost under ₦50,000 per unit. Central to the system is a lightweight Temporal Convolutional Network (TCN) running on a Raspberry Pi Zero, which processes raw sensor readings to detect contamination anomalies in real time. To address the scarcity of labeled data in rural field settings, the model is trained using self-supervised learning, reducing dependence on large annotated datasets. Upon detecting an anomaly, the system triggers SMS-based alerts via a GSM module. The system is currently being evaluated on publicly available water quality datasets with simulated low-cost sensor noise profiles. The expected result is a robust, low-cost monitoring framework that puts real-time contamination detection in the hands of communities that need it most.