Engineering Manager Interview Experience

Classify product into hierarchical classes and generate attributes for products. There are 5k leaf classes and assume every product can be classified into a leaf class

I designed a deep learning based classifier using both text data (product tile, description etc) and image. I use a bert model to encoder product text information and an image encoder (vision transformer, CNN) to encode image and then concatenate them together and sent to a FNN for classification. Follow up question include: (1)5k classes? is that too many classes? I modified to do classification level by level since it is hierarchical. (2) what about missing image? (3) how to store the result? (4) I also briefly talk about using generative model and class label to generate attributes