Team Member Story - Vishwajeet | WonderBiz Technologies

By Vishwajeet Behera

“Their existing OCR is just not good enough. Accuracy is much lower than they’d like. Apparently, the algorithm is completely ignoring special characters. Bohot zyaada problems ho rahe hai kyuki it’s an integral functionality right. I know it’s a lot to ask but would you like to work on this Computer Vision POC?”

I looked back at Kaustubh Mhatre, our CEO, and answered “Okay, I will give it a shot”.

I had always been fond of Machine Learning since my college days. But so are most young developers, so what’s the catch here? The catch is that I had years of experience only in Mobile App Dev upto that point. From juggling Android and Hybrid mobile apps to pushing the whole productization of a Computer Vision project was a big, big challenge.

But proud to say, I’ve always been an efficient worker with the ability to connect dots quickly. Looking at my performance Kaustubh offered me this POC, because this was like a grey area the organization were eager to explore.

When we started we weren’t really aware of how this concept would evolve into an Industrial solution. We had direction upto one point, but the scope was very obscure. And if you have experience with SDLC, you’ll know that this is a risky challenge.

With Computer Vision, we were aware about how CNN generally works but there was much research to be done. We quickly learnt how CV works, but still only minimal info was available to connect CV concepts to the actual utilization needed by the client. We developed an OCR and what it did was basically extract info from a Cause and Effect table. Now you might ask that OCR is easily available in markets then why did we create one from scratch? Our client had limitations bringing in a market OCR and hence they were using their own OCR to predict characters and extract info.

The difference we brought was in terms of quality. We upgraded the OCR to a huge extent, and this was our first milestone - Developing our own Industrial OCR and using it to extract info from Cause-Effect-Action tables. So we would exactly replicate and assemble together components and their relationships in terms of Cause-Effect-Action.

You must have guessed it by now that the POC was successful and the clients were impressed with how we built the entire blueprint from scratch and a solution ready for productization in just under a year!

When the founders gave me the offer initially, they themselves guided and supported me throughout the learning process. On the technology, on how the approach should be. Directly involved in all stages. Because we were dealing with core ML, deep-diving into algorithms and complex mathematical functions. They gave me regular inputs on the various possibilities.

Once the OCR was created then it was my job on how to finally implement it in the use case. 6 months later upgraded the solution to inculcate Masked-RCNN.

Today, I’m leading a ML Team of 8 people. Whatever I learnt earlier, I share with the young developers in the team to drive them forward. I guide them so that they can convert CV concepts into Industrial solutions. Now with our team multiplying, we jump into new-new challenges regularly involving different problem statements.

The freedom and guidance at WonderBiz helped change the trajectory of my career. It was almost like I was back in college - driving to innovate with the right passion and mentorship. Yes, you need to be on your toes all the time. But the freedom to experiment your own ideas and latest algorithms? Very few companies come to mind!