I don’t think you would have to use AI to convert gcode into a 3d model. The way I would approach it is basically run the gcode and take all the 3-D coordinates you get for where to print, then maybe run it through a filter or some such to extract just the outermost wall passes for each layer. Stack those up, do some sort of automated linking to create your triangles, and ship it off to the recognition AI.

it’s possible you could run that on a printer processor but unlikely.