The liquid argon time projection chamber (LArTPC) is an imaging detector that can record charged particle trajectories at sub-millimeter spatial resolution with detailed calorimetric (energy deposit) information. It allows us to measure neutrino interactions with high precision, making it the detector of choice for the current and future accelerator neutrino experiments. The Short Baseline Neutrino (SBN) program employs three ≅100 ton LArTPC detectors to measure neutrino oscillation at short distance (≅1 km) and to provide the definitive measurement of the electron neutrino excess observed by the MiniBooNE experiment which could be an indication of new physics [1]. The Deep Underground Neutrino Experiment (DUNE) employs four 10 kiloton LArTPCs to measure neutrino oscillation parameters, including the CP-violating phase, at a long baseline (≅1300 km), as well as to search for new physics processes such as nucleon decay [2]. Despite the potential of this novel detector technology, however, data reconstruction and analysis techniques for large-scale LArTPC detectors (i.e. millions to billions of pixels) remain challenging after many years of software development effort. Recently we developed a large-scale, full data reconstruction chain using machine learning (ML) techniques as a common solution for SBN and DUNE [3]. The chain has shown a strong promise for efficiently identifying neutrinos, and supports the whole spectrum of physics analyses by producing reconstructed physics quantities at all levels. This project will demonstrate the scalability of the chain on a GPU-based High Performance Computing (HPC) system. The demonstration includes Hyper Parameter Optimization (HPO), implementation of the reconstruction chain for the data production of SBN and DUNE followed by physics studies, and the development of an online system that can run the reconstruction on-the-fly on live data taken by the experiments.