How neurons are generated and diversified has long puzzled developmental neurobiologists. One of the outstanding process is that it is challenging to observe the earlier and not-yet-diversified progenitors. To overcome this, we devised a neural network classifier that distinguishes neuron types by the genes they express. The classifier was then applied backwards in development: Late pupal neurons are classified according to what we know in adult, and earlier neurons are classified with the later ones. By doing so, we were able to annotate every neuron that we profiled throughout pupation. With the ability to classify pupal neurons, we identified a neuron type that is only present in the pupal stages. They decorate the margins of the medulla, and express distinct Wnt depending on whether they are on the ventral or dorsal margin, which lead us to speculate they play a role reminiscent to Cajal-Retzius cells in mammals. Additionally, we also noticed that even though optic neurons are generated over a long period spanning several days, after 20 hours after pupation, their gene expression becomes synchronized with neurons of the same type despite their ‘age’ difference.