Accurate volumetric flow measurements are essential for many applications, such as power plant controls and district heating systems. However, flow metering under disturbed flow conditions, such as downstream of bends, can lead to errors of up to 20%. This repository contains an algorithm based on a shallow neural network (SNN) that significantly reduces errors for strongly disturbed flow profiles. The algorithm takes a velocity profile measurement along a 1D line as input and predicts a 2D profile, which is then integrated to determine the volumetric flow rate. The advantage of this general approach is that it can operate independently of metadata and be tuned using specialised training datasets. We used a dataset consisting of flow profiles behind a single 90° elbow (SE), a double elbow out-of-plane (DE) and a double elbow in an S-shape (DSE). On a test set, the algorithm achieved an average error of 0.42%, whereas the state-of-the-art method (assuming rotational symmetry of the 1D path) achieved 2.25%. In 83% of cases, the SNN produced a smaller error than the state-of-the-art method.