AAODP pt. 1 — Project structure and Milk & Cartoons Team Role

Published on Nikolay Aleshkovskiy

⚠️ The project is being developed under NDA with the client. By agreement and for security reasons we can’t reveal its name and owner. To present the product, all logos, labels, names and contents have been replaced with dummy data.

According to its purpose, we’ll call the project AAODP (Automated AI OCR document processing)

Milk & Cartoons Team Role

Our agency was originally hired a few years ago to implement only the front-end part of the project.

Automated AI OCR document processing application architecture diagram

Since then, we have taken a lead on a “larger front-end”, marked with a blue circle on a diagram below, which actually includes full-stack Meteor applications built entirely by our team.

Client Meteor Application

Each incoming documents goes through 4 main stages:

  • Uploaded/Received

    After a document is placed in an S3 bucket, it is processed by the OCR engine. A document type is assigned according to rules defined by business needs. This can be done automatically based on document content or via UI if no matching document type is found. UI allows to create new document types and rules to auto-assign them in future.

  • Classified

    At this stage, document type is set (automatically or manually) so that proper extraction rules are applied. AAODP looks for rules defined in the DB, extracts data from the document, processes data if necessary, and puts it into the UI for user review.

  • Extracted

    Depending on the application settings, the document can be automatically advanced to the next (archived) state or wait for the user to review and correct the extraction results via the user interface.

  • Archived

    Extracted data is sent to business information system, document marked as processed and archived.

© 2024 Nikolay Aleshkovskiy