At Greyparrot.ai, I helped build a world-class waste analytics platform. I led the Deep Learning team to create a system that automatically recognizes all trash items going through recycling facilities, helping recover all valuable materials and converting them to assets. Cameras were installed on top of conveyor belts that transport waste. Our tech accurately recognized around 50 distinct object categories and hundreds of brands, empowering robotic manipulation and data-driven insights.

As Head of the Deep Learning team, I managed the end-to-end process – from image intake to model creation, evaluation and deployment. I helped design and implement this process such that the full pipeline is done continuously with as little human input as possible. During my position, we had recognized billions of objects (yes, we humans create a lot of trash!), and we were able to recognize them with less than 5% error rate on each waste stream.

Here’s a demo of our waste recognition system:

One important challenge we overcame was reducing the need of massive amounts of high-quality annotated data for our Deep Learning models to work well. We developed an algorithm that helps the neural network identify objects that should be added to training set, so that only the most informative data is annotated. This helped us reduce annotation efforts by 100x.

I was very proud to be able to drive such an important technology to help reduce our waste and footprint on the planet.

Here’s a video where I explain this work further

Also, here’s a blog post of what a “Day in the Life” was for me as the Head of the Deep Learning team.

And don’t forget to heck out the amazing work at Greyparrot.ai