Advanced RPA Solution for Logistics Efficiency

Project background

In the post-pandemic era, even organizations that were slow to embrace automation are turning to RPA software as a quick and easy way to become more competitive and resilient without worrying about a big software project or system overhaul.

In this RPA for logistics case study, you’ll find out how our client can automatically identify such documents as invoices, bills of lading (BOL), proofs of delivery (POD), carrier supplier confirmations, letters of carrier assignment (LOA), and letters of release (LOR), and extract the following information from them:

– Invoice number​

– Invoice amount​

– Invoice date​

– LDS number​

– Supplier name​

– Remit to​

– QR code​

– Logo

Business needs

Every set of docs usually has from 10 up to 25 documents in different formats and languages. The customer company had a huge human staff to process these documents manually. In 2020, when the pandemic started, it became a necessity to reduce human staff to reasonable numbers and cut business costs with the help of custom RPA tools for logistics operations.

Our client’s goals included the following:

– To improve the logistics document process, automate manual tasks, and reduce human staff with robotic process automation in logistics.

– To check the idea of automated document processing and find a good development team.

– To increase the document processing speed. 

– To build an RPA platform for logistics capable of self-training.

Product features

1. Dashboard. Uploading PDFs on the Dashboard page
2. Automatic pages identification. Automatically identifying pages into the following categories: invoice, bill of lading (BOL), proof of delivery (POD), carrier supplier confirmation, letter of carrier assignment (LOA), and letter of release (LOR)
3. Information extraction. Extracting information from Invoices, BOL and POD identification
4. Actions with documents. Documents classification, required fields recognition, and text extraction
5. Detection and recognition. QR code recognition and reading, logo detection and recognition
6. Ability to learn for system. We gave our system the ability to learn when adding previously unknown types of documents and fields

Solution

CHI Software delivered an RPA solution based on machine learning, natural language processing, computer vision, and optical character recognition. This solution replaces manual human labor in the general workflow.

– The RPA process automation system allows receiving separate incoming logistics documents, upload one-page documents as scanned images, determine their type by checking the established set, extract text from predefined page fields, detect the signatures (if they are provided), recognize the QR codes, and compare data entered earlier in their system. In this case, the system returns them with a request/response, through the API.

– This RPA robotic process automation solution also allows our client to expand existing data (add new documents and recognizable fields to them, as well as edit existing documents).

– We are planning to make an application for sorting images of documents and extracting information from them. We will also create a service, hosted in Azure, for automatic reading and processing of email messages and a web application for sorting the results of processing.

NLP stack: 

– Transformers (XLM-RoBERTa, BERT) 

– Custom training of transformer models with a custom loss function 

– Embedding comparison algorithms 

– Dimension reduction 

– Data cleaning 

DRM stack:  

– Virtual Machine Service (VM)   

– Blob Storage 

– Azure SQL Database  

– App Service  

– Service Bus  

– Azure Functions 

CV & OCR stack (Torch, OpenCV, Sklearn):​

– Tesseract OCR   ​

– CRAFT (Text Detection model)​

– Yolo (text-blocks detection)

– Custom text blocks aggregation algorithm ​

– General Image Processing​

– QR code recognition

Our technology stack

• Python
• Angular
• OpenCV
• Tensorflow
• PyTorch
• NumPy
• Neural Networks

Key achievements delivered

1. We are currently working on an MVP development of the custom RPA solution for logistics that partially replaces manual human labor.
2. We conducted a profound discovery stage with several sessions and delivered a POC version.
3. The CHI development team trained the neural network to recognize and sort the defined list of documents.
4. We created a system that can be easily updated with new document field or type.
5. We made an ML pipeline both on servers and in the cloud.