isn’t dying, however it’s evolving. Quick.
AI-related jobs are projected to develop ~40% year-over-year, creating over 1,000,000 new roles by 2027.
On this article, I’ll take you thru my transition from Knowledge Science to Ai Engineering, in addition to provide you with some sensible recommendation on tips on how to transition or to be taught extra about this space.
My path by Knowledge Science to AI Engineering has been attention-grabbing and filled with learnings. Here’s a brief snapshot of my journey thus far:
- I graduated from Physics and Astrophysics (bachelor’s and grasp’s) and transitioned to Knowledge Science;
- Carried out two internships overseas in Knowledge Science and Machine Studying;
- Received my first full-time job as a Knowledge Scientist within the greatest power firm of my nation;
- Transitioned to AI Engineering lower than a yr in the past (as of Could 2025) and now I work for an enormous logistics firm.
If you’re a information scientist, how usually do you consider how your code reaches manufacturing? If the reply is ‘virtually by no means’, AI Engineering would possibly shock you.
Inquisitive about how real-world expertise in information science may form your journey into AI engineering, or what stunning challenges I confronted?
How is a every day lifetime of an AI Engineer examine to a Knowledge Scientist’s one?
What instruments and platforms I take advantage of now, in comparison with earlier than?
Maintain studying it to know all about it!
Howdy there!
My identify is Sara Nóbrega, and I’m a an AI Engineer.
I write about information science, Artificial Intelligence, and information science profession recommendation. Be sure to follow me to obtain updates when the subsequent article is revealed!
Variations and Similarities Between Knowledge Science and AI Engineering
AI Engineering is a really broad time period and it could even embrace many Data Science duties. In actual fact, it’s usually used as an umbrella time period.
As a Knowledge Scientist, I as soon as spent 3 weeks tuning a mannequin offline. Now, as an AI Engineer, we now have 3 days to deploy it into manufacturing. Priorities shifted quick!
However does that imply that each roles are fully completely different and by no means overlap?
What if sooner or later you wish to apply to an AI Engineer position? Are information science expertise transferable to the world of AI Engineering?
First, I’ll present you some findings of the analysis I did on this after which my private consumption and expertise on the topic.
I did a little analysis for you…
From my investigation, the obligations of every position have broadened and converged over the previous three years.
Knowledge Scientist job descriptions at the moment embrace increasingly duties apart from evaluation and mannequin tuning. They usually embrace: deploying fashions, constructing information pipelines, and making use of Machine Studying Operations (MLOps) finest practices.
Guess what, that is what I primarily do as an AI Engineer! (Extra on this within the subsequent sections).
For instance, a latest Knowledge Scientist posting I noticed explicitly required “expertise with enterprise DataOps, DevSecOps, and MLOps”.
Till some years in the past, information scientists centered primarily on analysis and modeling. Now, corporations usually count on information scientists to be “full stack”, which overwhelmingly means, fluent in virtually every little thing.
Because of this it’s anticipated that information scientists have some data of cloud platforms, software program engineering, and even DevOps, so their fashions can immediately assist merchandise.
One survey discovered 69% of information scientist job listings request machine studying expertise and about 19% point out NLP, up from simply 5% a yr prior.
Cloud computing expertise (AWS, Azure) and deep studying frameworks (TensorFlow/PyTorch) now seem in ~10–15% of information scientist adverts as properly, indicating a rising overlap with AI engineering ability units.
There’s a clear convergence within the ability units of Knowledge Scientists and AI Engineers. Each roles closely use programming (particularly Python) and information expertise (SQL), and each want understanding of machine studying algorithms.
In accordance with an evaluation of 2024 job postings, Python is required in ~56–57% of each information scientist and ML engineer listings.
Cloud and MLOps expertise appear to be the new widespread floor, as AI Engineers are anticipated to deploy on AWS/Azure and likewise “cloud expertise shall be important” for future information scientists.
The desk under highlights some core expertise and the way steadily they seem in job adverts for every position, in keeping with the sources that I record within the references part:
At first look, the divergence is apparent. Knowledge Scientist roles stay grounded in conventional information duties: Python, SQL, basic machine studying, and deriving insights from structured information.
ML/AI Engineers are positioned a lot nearer to the world of software program engineering. These professionals are tasked with taking experimental fashions and making them strong, scalable, and repeatedly deliverable.
However there’s a clear convergence that’s attention-grabbing and strategic.
We will see that cloud platforms are more and more talked about for Knowledge Scientists, and MLOps instruments are now not confined to engineering roles. The ability units are mixing!
We’re seeing a pattern the place Knowledge Scientists are being nudged nearer to the engineering stack.
My Private Journey and Consumption
What did I do as a Knowledge Scientist? What do I do know as an AI Engineering?
To present you some context, I labored as an information scientist in an enormous power firm, the place my obligations revolved round creating time-series forecasting fashions (utilizing XGBoost, LightGBM, SARIMAX, and RNNs), producing and validating artificial information (by way of TimeGAN, statistical distributions, and imputation methods), doing deep and intensive statistical analyses and using machine studying fashions to sort out lacking information in huge information.
If you’re , I wrote a ton of useful articles to cope with time-series information.
Among the instruments and platforms I used as a Knowledge Scientist included: VSCode, Jupyter, MLflow, Flask, FastAPI, and Python libraries equivalent to TensorFlow, scikit-learn, pandas, NumPy, Matplotlib, Seaborn, ydata-synthetic, statsmodels, and others.
In my earlier internship, I might use PyTorch, Transformers, Weights & Biases, Git, and Python libraries for information distillation, supervised studying, utilized statistics, laptop imaginative and prescient, NLP, object detection, information augmentation, and deep studying.
The instruments and platforms I take advantage of now
Python continues to be the principle language I take advantage of. I do use Jupyter notebooks for prototyping, however most of my time is now spent writing Python code in VSCode (scripts, APIs, checks, and many others).
My work could be very linked to Microsoft Azure, notably Azure Machine Studying, as my staff makes use of it to handle, prepare, deploy, and monitor our ML fashions.
Your complete MLOps lifecycle (from improvement all the way in which to deployment) runs in Azure. We additionally make the most of MLflow to trace experiments, examine completely different fashions and parameters and register all of the mannequin variations.
A significant shift for me from DS to AI Engineering has been the constant use of CI/CD instruments, particularly GitHub Actions. This was really one in all my first duties once I began this job!
GitHub Actions enable me to construct automated workflows that check and deploy ML fashions, in order that they are often built-in into different pipelines.
Past machine studying, I additionally construct and deploy backend parts. For that, I work with REST APIs, with FastAPI and Azure Features, to serve mannequin predictions and join them to our frontend purposes or exterior companies.
I’ve began working with Snowflake to discover and remodel structured datasets utilizing SQL.
Concerning infrastructure as a code, I’ve used Terraform to handle cloud sources as code.
Different instruments I take advantage of embrace Git, Bash, and Linux surroundings. These are essential for collaboration, scripting automation, troubleshooting, and managing deployments.
Some duties I’ve carried out as an AI Engineer
Now, I work as an AI Engineer for an enormous logistics firm.
The primary process I used to be assigned to was to enhance and optimize steady integration/steady deployment (CI/CD) pipelines of ML fashions utilizing GitHub Actions and Azure Machine Studying.
What does this imply in follow, you ask?
Nicely, my firm wished a reusable MLOps template that new tasks may undertake with out ranging from scratch. This template is sort of a starter pack. It’s in a GitHub repo and has every little thing you’d must go from a prototype in a pocket book to one thing that may really run in manufacturing.
Inside this repo, there’s a Makefile (a script that allows you to run setup duties like putting in packages or working checks with a single command), a CI workflow written in YAML (a file that defines precisely what occurs each time somebody pushes new code, for instance, checks are run, and fashions get evaluated), and unit checks for each the Python scripts and the configuration recordsdata (to verify every little thing behaves as anticipated and nothing breaks with out us noticing).
For those who want to be taught extra about this, I really wrote a full Dev Checklist for ML projects that describe these finest practices, and that’s completely beginner-friendly.
From linting and Makefiles to GitHub Actions and department safety, it’s filled with the sensible steps that I want I knew earlier:
👉 Read it here: From Notebook to Production — A Dev Checklist for ML Projects
Unit checks are literally a core a part of AI Engineering. They’re usually not the favourite process of anybody… however they’re essential for ensuring issues don’t break when your mannequin hits the actual world.
As a result of think about you’ve spent days coaching a mannequin, solely to have a tiny bug in your preprocessing script mess every little thing up in manufacturing. Unit checks assist catch these silent failures early!
However does this imply I’ve stopped performing Knowledge Science duties? Under no circumstances!
In actual fact, one in all my present duties entails mapping departure and arrival occasions, cleansing route information, and integrating the outcomes right into a frontend app.
I believe it’s a good instance of how Knowledge Science (EDA, mapping, cleansing) blends with AI Engineering (integration, deployment consciousness).
I wish to spotlight that each roles (Knowledge Scientist and AI Engineer) could be fairly broad and their obligations usually range from firm to firm, even sector to sector. What I’m sharing right here is simply primarily based on my private expertise, which can not mirror everybody’s journey or expectations in these roles!
Collaboration Patterns
One factor I’ve seen is that this overlap in obligations has compelled nearer collaboration with different staff members. I’ve seen that information scientists are more and more working side-by-side with DevOps and backend engineers to make sure fashions really run in manufacturing.
A study found that 87% of machine studying options fail to make it out of the lab with out groups coordinating in an environment friendly approach.
During the last years, corporations have acknowledged the necessity for collaboration. In actual fact, the necessity for MLOps finest practices have come to life to bridge this hole between information scientists and DevOps.
Largest Challenges So Far
I’m not gonna lie, this journey has been difficult. Everybody should pay attention to the imposter’s syndrome, and I’ve definitely suffered from it as properly. I assume it disappears over time as I really feel I add worth to the tasks I take part in.
Proper once I began to work as an AI Engineer, the greatest problem was to get used to new instruments, and to make use of all of them collectively. As I used to be assigned an essential process that solely I used to be engaged on (the MLOps template one), I felt I had instantly loads of accountability. I needed to shortly be taught the YAML language, Github Actions and the way they hook up with Azure.
Since I used to be actually into MLOps, I ended up taking up the position of system architect in just a few tasks. I used to be chargeable for determining how all of the items would match and work collectively, after which explaining it clearly to my managers.
I used to be undoubtedly not used to those obligations and roles, however over time I’ve grown extra assured in dealing with them.
Tricks to transition from DS to AI Engineering
I might say that step one to turn into an AI Engineer is to begin by being and curious about how the massive image of AI works. That is how I began.
That is how I began!
I began by asking myself: How will this mannequin really go reside to the customers? How will it add worth? How does the databases work, and the way can we fetch the info in manufacturing? How can I ensure that in 6 months this mannequin nonetheless works? How can I ensure that my mannequin shall be as correct regionally as in manufacturing?
Then, I began studying articles on-line and LinkedIn posts as properly, earlier than I transitioned to AI Engineering.
There’s a enormous quantity of helpful content material on-line, totally free. I additionally began taking some on-line programs so my expertise turn into extra stable.
If you’re in an information science position, you could possibly ask your supervisor to begin contributing to manufacturing code in your staff, or to incorporate you within the conferences with the AI Engineers. From my expertise, managers all the time like staff that wish to be taught extra.
Then, you’ll be able to be taught on-line about GitHub Actions, Docker, and Azure/AWS. Find out about essential manufacturing metrics like latency, uptime, monitoring
This can be a very brief roadmap, I’ll go away the remainder of the information for the subsequent article 😉.
Closing Phrase
My Mindset Shifted: Why AI Engineers Should Suppose Like Devs
To transition to an AI Engineering position, you will need to take into consideration the huge image of a ML lifecycle: that’s, to verify the mannequin will really work, create impression and add worth to the corporate.
What does this imply?
It means taking into consideration, throughout the entire lifecycle, how the mannequin shall be built-in into real-world methods — how it is going to be deployed, monitored, scaled, and maintained over time.
It means pondering past notebooks and coaching accuracy, and asking questions like: The place will this mannequin run? How will we replace it safely? What occurs if the enter information shifts subsequent month?
For these getting into or transitioning throughout the AI area, bear in mind: you don’t must grasp every little thing, however you do want to grasp how your work suits into the bigger image of the ML lifecycle.
The deeper your empathy for the “different aspect” of the pipeline, the extra impression you’ll have.
As you seen by this text, transitioning to AI Engineering for me has been about working on, studying about and proudly owning your entire ML lifecycle, not simply the mannequin coaching.
In my previous position as an information scientist, I used to be performing conventional DS duties like EDA, anomaly detection, information wrangling, mannequin improvement and packaging. Certainly, it was immediately linked to what I had realized in college.
As an AI Engineer, I really feel my every day duties are a mix of each roles. I nonetheless discover and clear information, however I really feel I must suppose like a dev, so I’m positive the fashions work in manufacturing and maintained over time.
Positively one of many greatest mindset shifts was studying tips on how to ship code prepared for manufacturing and in addition to develop a mindset of automation: automate installations, testing, deployment, monitoring.
It has been an attention-grabbing journey thus far, that I intend to doc and share additional on.
Thanks for studying! Hope you discovered this publish helpful.
🔔 Yet another factor!
I additionally write a free publication, Sara’s AI Automation Digest, the place I share month-to-month insights, instruments, and behind-the-scenes takes on AI, automation, and the way it’s reworking the way in which we work.
Subscribe now and get entry to my FREE AI Tools Library — a curated Notion database of 20+ AI instruments with real-world use circumstances, options, and limitations.
I provide mentorship on profession progress and transition here.
If you wish to assist my work, you’ll be able to buy me my favorite coffee: a cappuccino. 😊
References
The Interview Query 2024 Data Science Report: The Rise of AI Jobs (Updated in 2024)
MLOps: Connecting Data Scientists and DevOps Teams
The Future of Data Science: Job Market Trends 2025–365 Data Science
