isn’t dying, however it’s evolving. Quick.
AI-related jobs are projected to develop ~40% year-over-year, creating over one million new roles by 2027.
On this article, I’ll take you thru my transition from Information Science to Ai Engineering, in addition to offer you some sensible recommendation on easy methods to transition or to study extra about this space.
My path by way of Information Science to AI Engineering has been fascinating and filled with learnings. Here’s a brief snapshot of my journey up to now:
- I graduated from Physics and Astrophysics (bachelor’s and grasp’s) and transitioned to Information Science;
- Carried out two internships overseas in Information Science and Machine Studying;
- Received my first full-time job as a Information Scientist within the greatest vitality firm of my nation;
- Transitioned to AI Engineering lower than a 12 months in the past (as of Could 2025) and now I work for a giant logistics firm.
In case you are a information scientist, how typically do you concentrate on how your code reaches manufacturing? If the reply is ‘nearly by no means’, AI Engineering would possibly shock you.
Interested in how real-world expertise in information science might form your journey into AI engineering, or what shocking challenges I confronted?
How is a every day lifetime of an AI Engineer evaluate to a Information Scientist’s one?
What instruments and platforms I exploit now, in comparison with earlier than?
Hold studying it to know all about it!
Hi there there!
My title is Sara Nóbrega, and I’m a an AI Engineer.
I write about information science, Artificial Intelligence, and information science profession recommendation. Be sure that to follow me to obtain updates when the subsequent article is revealed!
Variations and Similarities Between Information Science and AI Engineering
AI Engineering is a really broad time period and it might even embrace many Data Science duties. In actual fact, it’s typically used as an umbrella time period.
As a Information Scientist, I as soon as spent 3 weeks tuning a mannequin offline. Now, as an AI Engineer, we have now 3 days to deploy it into manufacturing. Priorities shifted quick!
However does that imply that each roles are fully totally different and by no means overlap?
What if sooner or later you need to apply to an AI Engineer function? Are information science abilities 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 function have broadened and converged over the previous three years.
Information Scientist job descriptions at present embrace increasingly more duties moreover evaluation and mannequin tuning. They typically 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 Information 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, firms typically count on information scientists to be “full stack”, which overwhelmingly means, fluent in nearly the whole lot.
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 straight help merchandise.
One survey discovered 69% of information scientist job listings request machine studying abilities and about 19% point out NLP, up from simply 5% a 12 months prior.
Cloud computing abilities (AWS, Azure) and deep studying frameworks (TensorFlow/PyTorch) now seem in ~10–15% of information scientist adverts as nicely, indicating a rising overlap with AI engineering ability units.
There’s a clear convergence within the ability units of Information Scientists and AI Engineers. Each roles closely use programming (particularly Python) and information abilities (SQL), and each want understanding of machine studying algorithms.
In line with an evaluation of 2024 job postings, Python is required in ~56–57% of each information scientist and ML engineer listings.
Cloud and MLOps abilities appear to be the new frequent floor, as AI Engineers are anticipated to deploy on AWS/Azure and likewise “cloud abilities will likely be important” for future information scientists.
The desk under highlights some core abilities and the way continuously they seem in job adverts for every function, in accordance with the sources that I checklist within the references part:
At first look, the divergence is apparent. Information Scientist roles stay grounded in conventional information duties: Python, SQL, common 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 fascinating and strategic.
We are able to see that cloud platforms are more and more talked about for Information Scientists, and MLOps instruments are now not confined to engineering roles. The ability units are mixing!
We’re seeing a development the place Information Scientists are being nudged nearer to the engineering stack.
My Private Journey and Consumption
What did I do as a Information Scientist? What do I do know as an AI Engineering?
To provide you some context, I labored as a knowledge scientist in a giant vitality firm, the place my obligations revolved round growing 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 large information.
In case you are , I wrote a ton of useful articles to take care of time-series information.
A few of the instruments and platforms I used as a Information Scientist included: VSCode, Jupyter, MLflow, Flask, FastAPI, and Python libraries resembling TensorFlow, scikit-learn, pandas, NumPy, Matplotlib, Seaborn, ydata-synthetic, statsmodels, and others.
In my earlier internship, I’d 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 exploit now
Python continues to be the principle language I exploit. I do use Jupyter notebooks for prototyping, however most of my time is now spent writing Python code in VSCode (scripts, APIs, checks, and so forth).
My work may be very related to Microsoft Azure, significantly Azure Machine Studying, as my group makes use of it to handle, practice, deploy, and monitor our ML fashions.
The complete MLOps lifecycle (from growth all the way in which to deployment) runs in Azure. We additionally make the most of MLflow to trace experiments, evaluate totally 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 considered one of my first duties after I began this job!
GitHub Actions permit me to construct automated workflows that take a look at and deploy ML fashions, in order that they are often built-in into different pipelines.
Past machine studying, I additionally construct and deploy backend elements. 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 providers.
I’ve began working with Snowflake to discover and remodel structured datasets utilizing SQL.
Relating to infrastructure as a code, I’ve used Terraform to handle cloud assets as code.
Different instruments I exploit embrace Git, Bash, and Linux atmosphere. 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 a giant 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 apply, you ask?
Effectively, my firm needed a reusable MLOps template that new tasks might undertake with out ranging from scratch. This template is sort of a starter pack. It’s in a GitHub repo and has the whole lot 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 permits you to run setup duties like putting in packages or operating 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 information (to verify the whole lot behaves as anticipated and nothing breaks with out us noticing).
Should you want to study 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 typically 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 the whole lot up in manufacturing. Unit checks assist catch these silent failures early!
However does this imply I’ve stopped performing Information Science duties? Under no circumstances!
In actual fact, considered one of 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 excellent instance of how Information Science (EDA, mapping, cleansing) blends with AI Engineering (integration, deployment consciousness).
I need to spotlight that each roles (Information Scientist and AI Engineer) could be fairly broad and their obligations typically fluctuate from firm to firm, even sector to sector. What I’m sharing right here is barely primarily based on my private expertise, which can not replicate everybody’s journey or expectations in these roles!
Collaboration Patterns
One factor I’ve observed is that this overlap in obligations has pressured nearer collaboration with different group members. I’ve observed 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.
Over the past years, firms 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 concentrate on the imposter’s syndrome, and I’ve actually suffered from it as nicely. I assume it disappears over time as I really feel I add worth to the tasks I take part in.
Proper after 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 all of the sudden numerous duty. I needed to shortly study 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 over the function of system architect in a couple of tasks. I used to be answerable 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’d say that step one to develop into an AI Engineer is to start out 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 information in manufacturing? How can I ensure that in 6 months this mannequin nonetheless works? How can I ensure that my mannequin will likely be as correct regionally as in manufacturing?
Then, I began studying articles on-line and LinkedIn posts as nicely, earlier than I transitioned to AI Engineering.
There’s a big quantity of helpful content material on-line, without spending a dime. I additionally began taking some on-line programs so my abilities develop into extra stable.
In case you are in a knowledge science function, you can ask your supervisor to start out contributing to manufacturing code in your group, or to incorporate you within the conferences with the AI Engineers. From my expertise, managers at all times like staff that need to study extra.
Then, you’ll be able to study on-line about GitHub Actions, Docker, and Azure/AWS. Study essential manufacturing metrics like latency, uptime, monitoring
It is a very brief roadmap, I’ll go away the remainder of the information for the subsequent article 😉.
Remaining Phrase
My Mindset Shifted: Why AI Engineers Should Suppose Like Devs
To transition to an AI Engineering function, you will need to take into consideration the large image of a ML lifecycle: that’s, to verify the mannequin will really work, create affect and add worth to the corporate.
What does this imply?
It means bearing in mind, throughout the entire lifecycle, how the mannequin will likely be built-in into real-world programs — how it is going to be deployed, monitored, scaled, and maintained over time.
It means considering past notebooks and coaching accuracy, and asking questions like: The place will this mannequin run? How can we replace it safely? What occurs if the enter information shifts subsequent month?
For these coming into or transitioning inside the AI area, bear in mind: you don’t must grasp the whole lot, however you do want to grasp how your work matches into the bigger image of the ML lifecycle.
The deeper your empathy for the “different aspect” of the pipeline, the extra affect you’ll have.
As you observed by way of 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 function as a knowledge scientist, I used to be performing conventional DS duties like EDA, anomaly detection, information wrangling, mannequin growth and packaging. Certainly, it was straight related to what I had discovered 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 assume like a dev, so I’m certain the fashions work in manufacturing and maintained over time.
Positively one of many greatest mindset shifts was studying easy methods to ship code prepared for manufacturing and likewise to develop a mindset of automation: automate installations, testing, deployment, monitoring.
It has been an fascinating journey up to now, that I intend to doc and share additional on.
Thanks for studying! Hope you discovered this publish helpful.
🔔 Another factor!
I additionally write a free e-newsletter, 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 instances, options, and limitations.
I provide mentorship on profession progress and transition here.
If you wish to help 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
