Recently I’ve become interested in AI Engineering as a discipline. As an infrastructure guy, data science and analysis has always fascinated me, but I felt woefully unprepared to work with it. When I was in college, I took a course on relational databases and it left my head spinning. Building out a relational table design, worrying about normal forms, and writing SQL queries was not my forte.
However, I have to reckon with the fact we live in a world awash with data. Being able to wrangle the data monster is an essential skill for any information worker, and I suppose that includes me. AI Engineering is a new-ish discipline in the realm of data science that focuses on solving practical problems through the use of AI models. Given it’s practical nature, AI engineering holds an attraction for me that more theoretical disciplines do not. What can I say? I like seeing where the rubber meets the road.
But where to start? Like I said, I’m not a data scientist. I’m not deeply steeped in multi-dimensional data cubes, data lakes, or data warehouses. I don’t live in Jupyter notebooks or spend days tinkering with Amazon Sagemaker. How can I get started in AI engineering and what do I need to know?
When in doubt, ask an expert, and that’s exactly what I did. In this episode of Day Two DevOps, Kyler and I talk to Marina Wyss about AI Engineering. Marina is a Senior Applied Scientist over at Twitch, and she’s well versed in the world of AI engineering and machine learning. She helped me understand what an AI engineer does and the challenges they face.
Marina frames AI engineering as a distinct role from traditional machine learning or data science. Drawing on Chip Huyen’s definition, she explains that AI engineers primarily build products using pretrained models, often accessed via APIs, rather than training models from scratch. The core value of the role is not inventing new models, but selecting, adapting, and operationalizing existing ones to solve real product problems.
Where machine learning engineers and data scientists focus on data collection, labeling, and training custom models, AI engineers focus on:
AI engineers need to know enough about how models are developed and trained to leverage them successfully, but they aren’t going to be doing the actual training themselves. They need to understand the problem space, the goals of the applications, and what inputs will be available. Based on that information, they can take an off-the-shelf model and tweak it using some combination of RAG, fine-tuning, and prompt engineering to adapt it for a particular application.
One of the things I loved about the episode was discovering tools I’ve never heard of before. I live in a world of Terraform, Kubernetes, and YAML. Marina casually brought up Airflow and Dagster and I had to stop her and ask what those things are- turns out they’re tools to manage data pipeline orchestration. It would appear I’ve got some reading and experimenting to do.
Beyond the pipeline orchestration tools, she also stressed the need to be comfortable in Python and SQL, along with the public cloud offerings around machine learning. Since Twitch is part of Amazon, she is using AWS Sagemaker for a lot of her work, but really any of the major public clouds have ML tools as a service.
Learning how to use these tools effectively will require a ton of work on my part, and that’s something else Marina stressed. She didn’t become an AI engineer overnight, it was a continual process that involved self-study and off-hours experimentation. We didn’t get a chance to cover it, but Marina has a couple of excellent videos that review AI engineering courses and good starting points for folks of different backgrounds. Fortunately, I’m not start from scratch. I have a CS degree and a good grasp on data fundamentals, what’s missing is training on machine learning and model tweaking.
I really want to dig deeper into the world of AI engineering and build something in the process. I’ve started my journey by taking the Microsoft Learn courses around their AI-900 certification. Once I’m done that, I plan to start studying for the AI-102 certification for AI engineers. I don’t always enjoy certifications as a way to learn, but in this case I believe it’s the right approach. Marina actually just released a video on AI certifications and she gets the value of certs spot-on. She actually recommends the AI-102 Azure AI Engineer Associate cert or the AWS and Google Cloud equivalents.
Once I get those two certifications, I plan to take one of the courses Marina recommended, preferably one that involves building out an actual application. I’ve always been a hands-on learner and I’ll be more invested if it’s a project I care about. Ultimately, I’m not looking to get a job as an AI Engineer, just expand my sphere of knowledge and potentially be able to provide training in the discipline sometime in the future. I do have my MCT now, and I bet that the courses teaching Azure AI engineering are in high demand.
