Two-phase flows are commonly noticed in nature, engineering systems, and wide range of industrial processes. These flows are characterized as highly non-linear advection-dominated problems. For these advection-dominated problems, linear compression methods (such as proper orthogonal decomposition and reduced basis method) are not suitable, as they exhibit slow Kolmogorov N-width decay, which leads to inefficient and inaccurate linear-projection based reduced order models. To overcome this issue, there are a few recent pre-processing techniques that can be used to transform the full-order solution manifold \cite {reiss2018shifted, papapicco2022neural, long2023novel}. Here, we use a neural-network based pre-processing technique \cite{papapicco2022neural} that automatically detects the optimal non-linear transformation of the full-order solution manifold by exploiting a deep-learning architecture. It consists of two neural networks, 1) ShiftNet, which finds the optimal shift for the full-order solution manifold, to accelerate the Kolmogorov N-width decay, and 2) InterpNet, which learns the reference configuration and able to reconstruct the shape of reference configuration for each shifted centroid distribution. In this work, we use this pre-processing technique to develop purely data-driven reduced order models for 1D traveling waves and 2D two-phase flow test cases.