Hi @papertxt! (Sorry for the mention, just making sure you can see this.)
I don't have direct ideas for projects, more suggestions on how to go about it.
As I had a final project for a ML course to do, the main obstacle we had to overcome for the idea we came up with was finding data for our models to train on! So one advice I have for you is to look for potential datasets that are of interest to you before setting in stone the idea of your project.
Even better: find a dataset you can work on and then draw an idea from there. I won't leave you empty-handed, here are top places to look for datasets:
- Kaggle: this one is pretty popular and you probably have heard of it. There are also competitions (with cash prizes) and if you participate I believe you can have access to a GPU with a better computational power than Colab. (But don't trust my word on this, I myself haven't taken part in competitions.)
- Papers With Code: so mainly a research one, that, as its name says, has the different papers and the code associated with (usually github repo made available by authors of the code). In there you'll find big important datasets and the publication that goes with the dataset. Also some leaderboards of the published models performing different tasks on the datasets. Those datasets are great to do a first training of your model if it's a completely novel one, but I don't believe you'll use the datasets there directly. You get to know the state-of-the-art models on the task you want to tackle though and that is important.
- UCI Machine Learning Repository: It's been a while since I visited the website (almost did not recognize it at first), but it's a pretty great website to check! You have data type, subject area and even the possibility of import to python directly (not for all datasets).
Overall I hope this will be helpful for a start. You can DM me if you want more ideas, though I do think it's also part of the project to come up with your own idea. ;) ML is vast and can be applied in so many domains, so it's also up to your preferences in the domain of application. I prefer working with images and in health so I like tumor segmentation tasks, prognosis prediction, diagnosis prediction, etc...
(I also don't really know what level for the project is expected from you, so if you have any more specific questions, my ask box is open/DMs are ok.)