OCR Document Management Solution for Logistics Leader

Client

A provider of freight brokerage services

Industry

Logistics & Transportation

Services

Software Development

Tech

Google Cloud, Python, RestAPI, JMagick, SageMaker

Challenge

Our client — a US-based provider of logistics services — was at a breaking point. The influx of documents from thousands of shippers and carriers was overwhelming their outdated, inefficient document management system. Manual processes used for cumbersome and time-consuming tasks like document indexing were holding them back. As the number of documents continued to rise, the urgency to switch to automated technology only grew stronger. The company turned to ITRex for ML development and computer vision services to build a proprietary OCR document management solution that would automate their processes and improve productivity.

Our task was to:

●

Audit the client’s database to identify and prioritize document types, as well as establish the number of unique clients using unique formats

●

Develop an OCR document management solution (functional and non-functional requirements, architecture, tech stack, vision)

●

Develop a group of custom ML algorithms for OCR and train the ML models

●

Build an OCR library and develop APIs for its integration

●

Develop a custom user interface and a public API to allow access from the client’s multiple internal systems

Solution

ITRex has delivered a proprietary OCR document management solution that can be accessed from the client’s internal systems through a custom user friendly interface. The tool is capable of recognizing almost 20 types of documents received from carriers and shippers, with some types generated by up to 6,000 unique customers using unique formats. Its recognition accuracy for high-priority documents has been improved to 95%-100% through training. Implemented as a microservice for easy integration into any workflow, the solution has the following capabilities:

●

Classifying the type and load ID of scanned documents, with the ML model trained on several hundred documents to scale to more than 1 million documents

●

Recognizing fields in high-priority documents to increase processing efficiency from two fields recognized manually to every field in a document. This capability can be extended with ease to any type of document through mapping

●

Recognizing handwritten text (signatures)

●

Aggregating fields to enable search using secondary attributes

Impact

●

A dramatic reduction in operational costs through automation of manual processes

●

Enhanced productivity with a greater speed and accuracy of document processing

●

Lower infrastructure costs compared to implementation of an off-the-shelf cloud-based system, as the company is running its own servers to host the proprietary solution

●

The foundations for implementing a big data/reporting platform