Ask HN: How to pivot to a Machine Learning engineer?

Software engineer here with 10+ YOE building data (mildly) intensive applications: mainly back-end development experience (from legacy to modern/cloud-native applications, brownfields and greenfields).

(1) is it wise to do this transition?

(2) has anyone else here in HN done it?

(3) how can I do it if my job has no ML in it?

Is there an ML engineering practice that isn't focused on building models but more on managing/deploying/scaling models? i.e. can I avoid learning all the maths underneath?

I think they call that focus "AI Engineer".

Edit: just realized you might also be thinking of "MLOps". See end of comment.

It's what I have been doing for the last two years but I refer to it as "software engineer with a recent focus on generative AI".

I also think AI Integration Engineer is good but I have only really seen AI Engineer.

The thing is, up until a few years ago, doing useful things with AI generally did require something more like machine learning knowledge.

But now that we have general purpose models like gpt-4o, Claude 3.5, LlaVA, etc., you can just do an API call and in a day or two have a more functional system than what a machine learning engineer may have spent months training a custom model on previously.

So that somewhat explains the confusion. I think it's best to just honest. Most applications actually don't need "real" ML knowledge or custom neural network architectures or training a model from scratch.

I do think that ML is a good field to be in if you have the patience to learn the math and about neural networks etc. But I think that is not what you are talking about. And the architectures and models are very general purpose so as I said you can work on many applications of ML now without having that background.

Go to the Anthropic or OpenAI documentation and copy paste their examples and try inserting some customization into the system message using an f-string.

I think MLOps is also a thing. Go to HuggingFace and RunPod and practice deploying models with Python. Also find some tutorials on LLM pre-training, fine-tuning, and evaluations. Check out Predibase.

A big thing right now I believe is Diffusion Transformers. If you can find some article explaining how to run a training job for that, you may be able to help people.

If you want to "cheat", check out replicate.com. cog could be useful for self hosting ML models outside of replicate.com also.

When people talk about working with ML requiring a lot of math, what do they mean? That is to say, I have a degree in electrical engineering so I learned a lot of math. However, my programming career hasn't required me to actually use it since I graduated.

So I understand/"know" a lot of math. However, it would be tough for me to build back up to the point where I can, for example, solve differential equations again.

In a manner of speaking: Does a career in ML require a strong understanding/knowledge of math or does it require you to be able to solve a lot of math?

I was a MLE Tech Lead at Snap and laid some of the foundations of the generative AI infra at Snap. I would highly recommend MLE route as a very rewarding career path.

This book is a very good introduction to designing Machine Learning Systems for production: https://www.amazon.com/Designing-Machine-Learning-Systems-Pr...

This blog by the same author is highly recommended as an intro into building production grade AI and ML systems: https://huyenchip.com/2023/04/11/llm-engineering.html

To summarize answers to your questions:

(1) Yes it is wise to do this transition especially at an inflection point in the zeitgeist of the times as now

(2) Yes

(3) See above for resources on how to get started and reach mastery in the craft of ML Engineering.

Great question!

1) Depend on your own goal. Its wise if you believe in the future, market appreciate ML eng higher than software eng (within same years exp)

2) Yes. I did (> 3 years now)

3) Start find the ML job

Is there an ML engineering practice that isn't focused on building models but more on managing/deploying/scaling models?

Yes. Companies who adopt ML at large scale usually need this.

i.e. can I avoid learning all the maths underneath?

Yes its possible as long as you are focusing on infrastructure. Eg: create/monitoring pipelines, models, etc.

Noted: While you finding job, try to take professional ML engineering certification/program (Google, AWS, Azure, etc) - they will provide you hands-on lab experience with some case studies. The cost is super cheap