I work with granite/marble slabs. Currently only people who are intimately familiar with the industry and have seen many many slabs can differentiate and remember the names of colors.

But there's no way for a consumer to know the name of the color. They often come to us with a picture of a granite color wanting something similar.

An ML project would be to correctly identify a granite/marble color based on a picture.

You can check out some colors here: https://www.stonelandinc.com/inventory/

Note that there are 100s (1000s?) of granite/marble colors.

I have experience developing ML and DL applications. This sounds interesting.

Can you share more info or share email? Also, I can collaborate with Op to ensure efforts are synchronized.

Hi, can you leave me your email please ?

The main thing with making small demo apps while still a student is the data. You need to have a source of data you can use and apply whatever techniques too.

Novel business problems are often still unsolved not because making a ML is hard but because getting the data is hard.

So for your problems either use easy to get data and apply a new angle on it, or spend some money on mechanical Turk etc to generate the data you need.

Optimizing marketing spend - you can show how a ML model will beat a conventional past performance regression algorithm by increasing sales, reducing cost, etc.