Documents Understanding Case Study

Ennova Research
3 min readNov 24, 2021


Bring Intelligence to Document Processing

AI, Big Data & Machine Learning


A multinational company operating in risk management, HR and insurance intermediation provides know-how and support by defining strategies of risk control and mitigation.

The company’s objective was to introduce an intelligent system that would increase the efficiency of the company’s liquidators by using an autonomous recognition system for documents related to medical refund requests.


Ennova Research’s DOC Understanding solution is an automatic recognition system of documents that improves the effectiveness and efficiency of processing paperwork by automating the analysis of documentation thanks to Artificial Intelligence (AI) technologies.

Our solution is based on modules and microservices. The entire process of processing paperwork is performed in the following four key modules:

DOWNLOADER: an independent job that runs periodically and downloads medical refund requests, including documents and metadata, from company systems via REST API and stores the files to an S3 Storage and related metadata in MongoDB database

WORKER: an intelligent process that activates itself when it has enough refund requests to elaborate. It processes files and the related metadata communicating with MongoDB, S3 Storage and Google Cloud Platform AI services.

For each refund request, the main tasks that are executed are:

  • pdf and image enhancement
  • image classification
  • OCR
  • NLP and entity extraction
  • object detection

PRESENTATION: an advanced web application whose purpose is to allow the company’s operator to validate potential AI failure or data missing match. It’s used by specific operators (usually referred as “validators”) that must supervise and fix (only if required) the output results of the Worker module. Administrator users can also verify and check each single refund request information.

UPLOADER: an independent job that runs periodically and uploads the validated and matched metadata to the original system via REST API


Baseline services:

  • S3 Storage (Minio): document storage
  • NoSql database (MongoDB): logical archive of the document metadata
  • Node-RED: workflow manager that processes the files and invokes the necessary microservices
  • Cache (MemCache): used to improve performance for visual presentations
  • Internal DB (Sql Server): interface for accessing databases
  • Active Directory: contains users and their roles
  • NodeJS + Express: backend that reads and writes in the No-SQL db and integrates with the customer services
  • Angular: draws the Web UI used by the validator

Google Services:

  • OCR: performs character recognition in PDF documents
  • Image Classification: custom AI model deployed in GCP AI Platform
  • Object Detector: custom AI model deployed in GCP AI Platform
  • AutoML: entity extraction from document text

How does it work?

A generic user owning a valid medical insurance policy uploads documents to a refund request collector’s portal, and thus initiates the refund process. Depending on the refund request type and the included medical service, the refund request is imported in the AI system through the Downloader module. The AI system performs the necessary optimization, verifies that the request’s files have not already been processed and then starts the elaboration process.

Each document is sent to Google Cloud Platform to apply the powerful optical character recognition (OCR) function and various AI techniques to get all the useful information that can be extracted from the document. Such information is then stored on MongoDB.

At this point, the elaboration result may be reviewed from a human operator due to missing data or automatically sent back to the original system because all the included information is matched. The eventual validation process serves to fuel the automatic learning by the system. At the end of this process, the document is filed as processed or incomplete (indicating the reason) and the outcome is sent to the original system.


As an immediate result, the client obtained an increase in efficiency by its liquidators with optimized human resource allocation and costs.

The documents and key information to be analyzed in liquidation cases were already highlighted, thereby resulting in an easy consultation by the liquidators.

The long term advantage is to automatically liquidate an increasing amount of cases by taking advantage of the continuously-growing and reliable AI technologies.

For more information about our solution, contact us at:

Discover more about Ennova Research



Ennova Research

We see beyond today’s needs focusing on R&D to bring innovative, human-driven and tech-driven solutions to the market.