The aging process or flow-induced vibration of reactor cores may lead to increased mechanical vibrations, affecting the reliability of in-core sensors and necessitating a robust solution for robust field reconstruction. This work tackles the challenges of reconstructing multiphysics fields from sparse and movable measurements by introducing an advanced framework that integrates various machine learning models with Voronoi tessellation. Our approach, building upon the Voronoi tessellation-assisted Convolutional Neural Network (VCNN), expands the capabilities to include a wider array of neural network architectures such as Convolutional Neural Networks (CNNs), Fourier Neural Operator (FNO), Dilated ResNet Encode-Process-Decode (DilResNet), Dilated Convolution Neural Operator (DCNO), Galerkin Transformer (GT), U-shaped Neural Operator (UNO), and Multiwavelet-based Operator (MWT). The effectiveness of these models is evaluated and validated through numerical tests based on the International Atomic Energy Agency benchmark, particularly noting average relative errors below 5% and 10% in the norm and norm, respectively, within a 5-cm amplitude around sensor nominal locations. The developed software toolkit encapsulates these architectures, providing a versatile option for nuclear engineers to reconstruct different types of physical fields efficiently.