Just keep working. Masters programs suck anyway --- the only things that really make a difference are a) time, b) conversations with peers, c) conversations with an advisor. When you're 22 a PhD is a reasonable deal to spend 4 years to get this. When you're 32 it's a significantly crappier bargain.
If you get good at designing the models, your results will speak for themselves, and it won't matter whether you have a certificate saying you can design models. Likewise if you're actually bad at something you're supposedly certified to do, I think that's actually worse than not having the qualification at all!
The simplest thing to do is just pick a problem with a decent dataset and work on it. Either build the model from scratch, or if you prefer to use a library, make sure you're compiling the library from source. You want to make sure you can change arbitrary things --- don't just use an API.
If you're interested in NLP, I can suggest a number of things you might work on for spaCy :). For instance, I'd like to have a sequence-to-sequence lemmatisation model. This will take some experimentation, but the basic idea should definitely work.
If you need any more convincing not to do a masters...Really think about what that will involve. You'll go to lectures. You'll get homework. You can do better than that.
If you get good at designing the models, your results will speak for themselves, and it won't matter whether you have a certificate saying you can design models. Likewise if you're actually bad at something you're supposedly certified to do, I think that's actually worse than not having the qualification at all!
The simplest thing to do is just pick a problem with a decent dataset and work on it. Either build the model from scratch, or if you prefer to use a library, make sure you're compiling the library from source. You want to make sure you can change arbitrary things --- don't just use an API.
If you're interested in NLP, I can suggest a number of things you might work on for spaCy :). For instance, I'd like to have a sequence-to-sequence lemmatisation model. This will take some experimentation, but the basic idea should definitely work.
If you need any more convincing not to do a masters...Really think about what that will involve. You'll go to lectures. You'll get homework. You can do better than that.