15/04/2026 5:10 pm
Topic starter
Notifications
Clear all
Agent Skills Forum
2
Posts
4
Users
0
Reactions
10
Views
0
So I'm currently working as a junior full stack dev in Chicago and I've been really thinking about switching over to AI engineering within the next 6 or 8 months. My math skills are pretty much stuck at high school level plus maybe a bit of logic from my CS degree, which honestly wasnt that math heavy to begin with. Ive been doing some digging online and the advice is all over the place which is making me kinda anxious about where to spend my time.
I read on some subreddits that if you dont understand multivariable calculus and linear algebra inside out you'll never survive a real interview at a place like Google or Meta but then I see these roadmap videos on YouTube that basically say you just need to know how to call APIs and use high-level libraries like Keras or PyTorch. It's super confusing because I dont want to waste three months studying derivatives if I'm just gonna be tuning hyperparameters all day. I have about $500 saved up for a bootcamp or some courses but I dont want to blow it on a theory-heavy thing if it's not actually what people use in the field every day. Like do you actually solve equations on the job or is that just for researchers? How much of that deep math do you actually use when youre building stuff for real clients...
12
15/04/2026 5:10 pm
Honestly, the massive gap between what companies ask for in interviews and what we actually do daily is pretty frustrating. I spent a fortune on high-end theory courses early on and frankly, it was mostly a waste for the engineering side of things. Most people unfortunately end up in one of two extremes: either they are just API wrappers who dont understand why their model is failing, or they are math wizards who cant write production-ready code. Based on my experience building production pipelines, here are two tips to keep you from wasting your $500:
- Prioritize Linear Algebra over Calculus. You rarely solve derivatives by hand because of autograd features, but you will constantly run into dimension mismatch errors. If you dont understand how matrices multiply, you wont be able to debug your architecture when using Meta PyTorch 2.2 Deep Learning Framework.
- Skip the generic AI bootcamps. They are usually overpriced and not as good as expected for the depth they provide. I had issues with the surface-level stuff in those. Instead, invest in O'Reilly Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow 3rd Edition and only dig into the math sections when you hit a specific wall. You dont need to be a researcher, but if you cant read a paper and understand the notation for a loss function, youll struggle to implement anything new. Focus on the why behind the functions, not just the how of the API.
