Okay so... You've obtained enough knowledge to know this stuff is possible. But you haven't obtained enough knowledge to know how big of a task this thing is. That is what I am trying to say.

Focus on doing one thing at a time.

My idea is to train a model myself through fine tuning, retraining it with adversarial deep learning.

You're already mixing up too many things. Training and fine-tuning are two different phases of model development.

After development, comes deployment, where you package your model into an application or a microservice. Either way, you need it to run somewhere. You can load in the model weights as a file anywhere, but just because you can have the file on disk, doesn't mean your devices will have RAM to load it.

You're trying to do too many things at once. You need to table like 90% of those ideas and focus on one thing at a time. To me it sounds like you barely know how to train a ML model. From what you're saying, you don't even have a dataset to train the model on.

Worry on generating a dataset first. Then worry about the quality of that dataset.

Then train a model with that dataset.

Then fine-tune it with some method of cross-validation. Move your biases, your learning rates, regularizations, etc.

Do you not know what these things are? Go study them.

This is already a good project in of itself. Don't bother trying to deploy your model. And by the way:

The idea is to see how I can run this model in different locations, such as on a desktop on my PC, on my phone, or perhaps connect it to an AWS cloud and see if it can read the traffic as well.

This is what deployment entails.

You seem to have a good high-level idea for your project, but there's a good bit of learning you have to do to connect the dots.

Literally ask AI to help guide you through studying this.