We made a computer vision solution for a client that allows dock workers to quickly scan codes from shipping containers into an inspection app on their mobile phones so they can inspect the freight contained in the shipping containers. What sounded like an easy OCR project turned out to be quite an undertaking. We learned a ton from this project. For example that building a deep learning model can be quite a challenge especially when ready-made models turn out not to work for your case.

In this session, we’ll talk you through our approach and the lessons we learned. We’ll cover the following topics:

- How we made an MLOps environment on Azure without spending too much.
- How we handled a dataset with 5.3 million samples and how that broke our environment.
- How we deployed our model to Android and IOS so it can use the model without an internet connection.
- How we’re using A/B testing on Android devices to find out which model is the best.

If you’re interested in learning about MLOps outside the lab then this is certainly going to be entertaining.