Hi all, i'm writing my bachelor thesis about "Processing of Invoices for the Extraction of Totals and VAT Information". The practical part should consist of a system which predicts the total sum, tax, ... for a provided image of an invoice (for the sake of simplicity the text is already extracted so no real text extraction is needed). Is KotlinDL suitable for this kind of Deep Learning resp. does anyone have an idea how to solve this? (Python is not my fav language thats why I try to avoid pytorch) Would be happy if someone could help me or writes to me directly.

At this moment, KotlinDL is not the best tool for training from scratch NLP models. If the text contains only numbers, the classical ML algorithms probably could be enough... If text is more complex, probably you need manually to extract some features from the text.