Machine Learning (supervised) Algorithm for automated processing of bank receipts via .pdf files
Budget: €5,000 – €10,000 EUR
We are looking for someone to develop machine learning algorithm, that recognizes atleast 8 different kinds of transactions and is then able to convert 5+ distinct informations to input strings for our web-based, self-developed input mask (see example).
So instead of having to manually view and process every single statement, our clerk would be able to load multiple .pdf files at once, attach them to a customer from our database, have them be processed automatically and receive notification about the statements that were processed incompletely.
The bank receipts can vary in layout and slightly in information description, e.g. a “buy – order” could be called “buy”, “purchase” or “acquisition” across different banks. We have atleast 50K+ examples of high-resolution singular transaction statements as .pdf at our disposal, distributed across 60+ different german banks. The algorithm doesn’t have to have a high hit-rate when delivered, if it is supervised-learning based as we continue to process new .pdf files via our input mask and “verify” new training data.
So instead of having to manually view and process every single statement, our clerk would be able to load multiple .pdf files at once, attach them to a customer from our database, have them be processed automatically and receive notification about the statements that were processed incompletely.
The bank receipts can vary in layout and slightly in information description, e.g. a “buy – order” could be called “buy”, “purchase” or “acquisition” across different banks. We have atleast 50K+ examples of high-resolution singular transaction statements as .pdf at our disposal, distributed across 60+ different german banks. The algorithm doesn’t have to have a high hit-rate when delivered, if it is supervised-learning based as we continue to process new .pdf files via our input mask and “verify” new training data.