In this paper, we present a novel non-invasive water flow metering technique that is cheap and exhibits decent performance. Targeting mainly irrigation monitoring, the technique has been applied to create a prototype measuring apparatus consisting of a small, battery operated board that includes both a vibration and an acceleration sensor. Data acquired from those sensors is then processed on-board via a neural-network that has been pre-trained and calibrated in the lab. The inferred water flow rate is then transmitted via LoRaWAN to a data back-end for further processing. With this device, we demonstrated that for a total cost of less than 18 C, our prototype communicating sensor could run for a complete irrigation season on 2 AAA batteries with data sent every 20 minutes. Regarding the performance of this AI-augmented sensor, the results exhibit less than 10% of error for most flow rates when compared to a fully calibrated, lab-grade water flow meter, with potential for improvement.