We present the results of applying a transformer-based deep learning neural network to the data from the HAWC Gamma-Ray Observatory. HAWC observes the extensive air showers produced by very high energy gamma rays, and registers the Cherenkov radiation produced by the shower by photo-multiplier tubes (PMTs) instrumented in 300 large water Cherenkov detectors. The current HAWC method uses a staged parameterized fitting of the PMT information to find the shower center and incoming angular direction of the initiating gamma-ray, and produces variables which can be used for separating showers produced by gamma rays versus the overwhelming cosmic-ray background. The deep learning model, on the other hand, takes the charge and relative timing information of the PMTs as input and directly outputs an estimate of the incoming direction of the initiating gamma ray and a gamma-hadron discriminator. Both tasks are vital for source analysis. Better angular reconstruction allows for better source localization. Improved cosmic-ray rejection improves the signal-to-noise ratio. The deep learning network is found to perform better in simulation than the current methods at lower energies (around several hundred GeV gamma rays) where fewer PMTs are turned on by the shower, and therefore less information is available.