The Liquid Argon Time Projection Chamber (LArTPC) technology enables unprecedented spatial and calorimetric resolution to observe intricate, multi-particle final states of neutrino interactions. Pandora provides a robust framework for reconstructing these complex event topologies observed in LArTPC experiments. An example of such a detector is MicroBooNE, which was designed with the primary goal of characterising the apparent low-energy excess (LEE) of electron neutrino events observed at MiniBooNE. With the first set of MicroBooNE LEE results now released, now is the crucial time to consider any feasible upgrades in the analyses before studying the remaining available data. This poster will describe the investigation into the potential impact of low-level reconstruction aspects in Pandora on the ability to select electron neutrino events and the sensitivity to the LEE signal. Motivated by the results of this study, it will also outline the performance of a novel deep-learning approach in Pandora for interaction vertex reconstruction.