Welcome to my new blog series where I'll be sharing details about our previous projects.To kick things off, let me tell you about a particularly exciting project we completed in 2019 for a well-known US-based procurement company. Our task was to automate all invoice and expense procurement tasks.
The company had a big idea: to automate all of their invoice and expense procurement tasks. This meant extracting data from uploaded invoices, including:
- Invoice dates
- Invoice numbers
- Total amounts
- Taxes
- Line item quantities
- Base and total prices
- Provider names and addresses
All of this had to be done without any human intervention.
Here's what we did:
Proof of Concept Application:
We provided a basic POC application that ran on Azure Functions to demonstrate the feasibility of the project.
Collaboration with a Distributed Team:
After the POC, we collaborated with a distributed team from Belarus to implement the final product.
Text Extraction API Testing:
We tested three different text extraction APIs from Azure, AWS, and Google Cloud, along with open source alternatives like Tesseract.
Performance Measurement:
We built an application to measure the performance of each solution by creating a sample test application. We curated over 100 invoices that contained specific edge cases.
Choosing the Best Solution:
After testing, we decided to go with Azure Cognitive Services as it gave the best performance on our training dataset.
Building A Microservices Architecture:
We developed a fully scalable microservices architecture over Azure Service Fabric. It included Elastic DB pools that scaled per tenant and Azure Service Bus for messaging between components.
Finalization:
The project was finalized in under 12 months and became a flagship product of the company. It was successfully sold the following year.
I hope you found this post informative and useful. Stay tuned for more interesting stories about our previous projects.