Fluorescence light detectors have been a crucial part of ultra-high energy cosmic ray observatories,
facilitating the study of the longitudinal development of extensive air showers. In this contribution,
we evaluate the feasibility of using neural networks to reconstruct shower geometry, using the
Fluorescence Detector at the Pierre Auger Observatory as a case study. We compare our results
to the standard reconstruction algorithm, assessing the performance of various neural network
architectures in reconstruction of key shower geometry observables. Additionally, we briefly
discuss the challenges in reconstructing the primary particle’s energy. Our results show that the
tested architectures do not outperform the standard method in the reconstruction of regular events
observed at the Pierre Auger Observatory.