Watatumi is an AI-powered software suite for monitoring critical fish farming activities, including feeding, cage monitoring, and fish transferring. It features deep neural network models for fish counting, fishnet damage detection, and fish size estimation. These models, powered by innovative efficient neural network architectures, deployed on embedded chipsets near underwater cameras, eliminating the need to transmit large video data out of water, improving reliability, reducing costs, and simplifying integration. The algorithms consistently achieve high accuracy under varying  lighting, water, and weather conditions, and adapt easily to different camera setups. Watatumi significantly improves the efficiency and profitability of fish farming.