Event reconstruction algorithms are used to infer the particle properties, such as energy and direction, based on the photosensor information. Traditional likelihood-based algorithms use several approximations in the modeling of the detector that limit its accuracy and speed, which must be improved for Hyper-K. Several algorithms (DNNs; ResNet CNN, GNN, PointNet, UNet) have been adapted to our particular data format and need to be applied to real physics data. Two positions are available for this project: a) application to CERN particle beam data in the Water Cherenkov Test Experiment, b) application to Super-K cosmic ray and atmospheric neutrino data.