This work explores machine learning methods for reactive transport at pore-scale, common in many industrial applications. Reactive flow in catalytic filters is described by a parametric convection-diffusion-reaction partial differential equation. The first part focuses on neural network methods for solving these equations, specifically physics-informed neural networks and a modified deep Ritz method. Improved performance is observed, but computation time remains a bottleneck. The second part examines surrogate models for reactive transport problems in porous media, relevant to fuel cells, photovoltaic cells, and catalytic filters. The efficiency of filtration processes is evaluated using breakthrough curves. Surrogate models predict these curves for new parameters, using data from numerical simulations of an artificial filter geometry. The predictions are accurate across different regimes and provide a significant speed-up in the parameter identification problem.