Our hybrid approach is well suited for computer vision projects.

One of our first computer vision problems involved real-time sausage detection in large supermarkets. Our system ensured that the grilled sausages were well cooked and optimized for the best-sellers.

How to follow in real time what happens with all the sausages?

Follow up very similar and small objects in real time without internet?

In ML there is this saying:

Garbage in, garbage out.

Not so fast. With our hybridization of ML and OR we built a system that takes bad inputs and still provide good insights!

In this computer vision problem, we had many small and identical objects (nothing looks more like a sausage than another similar sausage!) that we needed to follow upon in real time.

Also, all computations needed to be done on an edge computer with no internet access.

Moreover, the data was of poor quality (the damp on the cameras) and sometimes non existing (because of the heat, some cameras went off).

A pure ML/DL approach wasn't enough to successfully solve this problem! Our system was able to bypass the quality of the input and provide a robust solution in real time!