<p>Predicting heat transfer in variable porosity medium with the neural network approach allows for more precise predictions, which can result in better cooling, better lubrication systems and better design of biomedical devices. Keeping these important applications in view, the current work examines the thermally radiative flow of micropolar nanofluid on a variable Darcy regime through two circular disks. The main equations are first solved using the bvp4c approach and then the data from the bvp4c approach is utilized as a fundamental data for the artificial neural network (ANN) approach. This study shows that the ANN model exhibits good convergence behaviour, in the sense that for all of the operational cases a stable value of the MSE was reached within an acceptable range of 10⁻⁷ to 10⁻¹⁰. Error histogram analysis provides a small error uncertainty range, and more than 95% of error deviations are tightly grouped near the centre within a narrow absolute range of uncertainty of ± 5 × 10⁻⁶. The correlation between the predicted and numerical target values is outstanding for both fitness and regression analyses, both of which give a correlation coefficient of (<i>R</i> = 1.0000), indicating the high degree of statistical certainty of the predictive model. Increase in the Reynolds number and nanoparticles concentration results in increase in axial and tangential velocities and decrease in magnetic parameter results in decrease in axial and tangential velocities. Thermal profiles increase with the increase in Reynolds number and the radiation parameter and decrease with the increase in the heat source factor and the increase in nanoparticles concentration. Lastly, a quantitative validation of the comparison results with literature benchmarks reveals a very good agreement, with a maximum difference of less than 0.1%.</p>