Machine-learning models for image processing

A method for processing and validation of document imagery, the method comprising:

• obtaining, by a user device associated with an end-user, image data using an imaging software program locally executed on the user device, the image data comprising data depicting a document, and a label annotation indicating a user-entered quantity mapped to a quantity depicted by one or more fields of the document;
• identifying, by the user device, a type of the document from a plurality of predefined types;
• selecting, by the user device based on the type of the document, a retrieval-augmented generation (RAG) system from a plurality of RAGs for a large-language model (LLM) of a machine-learning architecture, the plurality of RAGs corresponding to the plurality of predefined types, wherein the selected RAG is configured to retrieve domain specific contextual data for the identified type of document and to provide the contextual data as input to the LLM;
• providing, by the user device, based on detecting a mismatch between the user-entered quantity and a detected quantity of the image data, the image data to the LLM;
• identifying, by the user device, an update to the detected quantity based on textual content received from the LLM;
• and
• validating, by the user device, the image data and the updated quantity for transmission based on identifying a match between the updated quantity and the user-entered quantity.