> Some teams build software with ML features. They may do some in-house research or reuse existing models/architectures. The ML feature can be a critical part of the product, or it can be a nice-to-have.

I am more interested in this. Understand from the perspective of building ML-based software applications/product-features (updated the question).