Career guide · 2026 · LU data science students
AI, ML, Data Science -
which career are you actually choosing?
You have probably noticed that "AI Engineer," "ML Engineer," and "Data Scientist" get used almost interchangeably in job posts and course descriptions. They are not the same thing. Choosing the wrong path means spending years building skills that do not match what you actually want to do. This guide cuts through that.
Part one
Before the roles, what do AI and ML actually mean?
The reason these career titles get confused is that the fields themselves get confused. AI and ML are not synonyms. They describe different scopes, and understanding that hierarchy is the foundation for everything else in this guide.
Any technology that makes a machine simulate human thinking or decision-making.
A way to build AI where the system learns patterns from data instead of following hand-written rules.
A type of machine learning built on neural networks with many layers, inspired by how the brain is structured.
Deep learning models trained to produce new content: text, images, code, audio. This is where LLMs like ChatGPT and Claude live.
The field of extracting meaning from data using statistics, ML, and domain expertise to support decisions. It extends beyond ML into business context and communication.
- Data Scientists work across the ML layer, using statistical models and data to drive decisions.
- ML Engineers go deeper, building and maintaining the systems that train and serve those models.
- AI Engineers sit at the very top - the generative AI layer - building products on top of foundation models without touching their internals.
The title "AI Engineer" sounds like the most advanced of the three - but it actually requires the least knowledge of how models work internally. That is not a weakness. The depth moves somewhere else: into system design, product thinking, and making AI reliable in the real world.
Part two
The three roles: what they are and what the work looks like
Each role has a core question it exists to answer. Flip the card to see a real scenario, and see where your time actually goes each week.
The analyst
"What is happening in this data, and what should we do about it?"The Data Scientist's job is to extract signal from data and translate it into decisions. This means designing experiments, building predictive models, and communicating findings to stakeholders who do not share your technical vocabulary.
What surprises most students: model building is a minority of the actual work. The majority is data wrangling, feature engineering, experiment design, and building the narrative around what the data says. Strong SQL and statistical reasoning matter as much as knowing how to fit a model.
An e-commerce company notices that sales dropped 20% last month. The manager asks: "What happened?"
"What is causing the drop, and what should we do about it?"You look at the data from different angles, by product, region, and time. The drop is concentrated in one category.
You trace when it started, and it lines up with a pricing change 3 days earlier.
You run a quick statistical test to confirm this is not just a coincidence.
Traffic is fine, but conversions dropped, people visited but didn't buy.
You write a short report with one clear message and recommend reverting the price or running a proper test first.
Where your time actually goes
Figuring out what the business actually needs to know and if data can even answer it.
Collecting, cleaning, and exploring data. Handling missing values and creating meaningful features.
Applying statistical methods and ML models. Interpreting results, not just optimizing performance.
Scientist
Evaluating models and turning results into actionable insights for decision-makers.
Presenting findings through dashboards, reports, and storytelling for non-technical stakeholders.
The math isn't the hardest part. The hardest part is convincing a business leader to actually change their strategy based on a chart they don't fully understand yet. Translation is your most valuable skill.
The model builder
"How do we turn raw data and a problem into a model that actually solves it?"The ML Engineer designs and trains machine learning models from the ground up. This means choosing the right architecture, understanding why the model is failing and fixing it, and evaluating performance against metrics that actually reflect the business goal.
Once the model is ready, the ML Engineer is also responsible for getting it into production, building training pipelines, setting up monitoring, and managing retraining when the model starts to degrade. This role requires deep mathematical understanding combined with software engineering discipline.
A bank wants to automatically flag fraudulent transactions before they go through.
"How do we build a model that catches fraud without blocking real users?"You study the data and spot a problem: fraud cases are extremely rare. A naive model would just call everything "normal."
You fix the imbalance and train a model that learns the actual patterns of fraudulent behavior.
You evaluate carefully: how often it misses real fraud vs. how often it wrongly blocks a legitimate user.
You keep adjusting until the balance between catching fraud and not annoying customers feels right.
You build a pipeline so the model retrains automatically every month as fraud patterns evolve.
Where your time actually goes
Translating business requirements into an ML problem. Deciding approach and defining real-world success metrics.
Designing reliable, scalable pipelines that ingest, clean, validate, and transform data into usable features.
Building and iterating on models. Hyperparameter tuning, versioning, tracking experiments for reproducibility.
Engineer
Evaluating models with realistic validation strategies. Optimizing for performance, latency, and cost.
Deploying models to production and continuously monitoring their performance over time.
The model is the easy part. The system is the hard part. Training in a notebook is straightforward. Making it work with messy, changing data, scaling it, and keeping it stable in production is where most of the work really is.
The product builder
"How do we build a working product on top of a pre-trained model?"The AI Engineer builds products and systems powered by AI, using LLMs, vision models, speech models, or any AI capability available through an API. The common thread is taking models that already exist and turning them into something useful for real users.
The hard part is not calling the API. The hard part is everything around it: designing the system, making sure the model gets the right context, handling failures gracefully, evaluating output quality, and keeping the whole thing reliable and cost-efficient when real users hit it.
A company has thousands of internal documents. Employees waste hours searching manually. They want a chat interface where employees can just ask a question and get an answer.
"How do we build something that finds the right answer without making things up?"You build a system that breaks all documents into smaller pieces and makes them searchable by meaning, not just keywords.
When a user asks a question, the system finds the most relevant pieces from the documents automatically.
Those pieces are passed to an AI model which reads them and writes a clear, direct answer.
Every answer includes a reference to the original document so users can verify it themselves.
You handle the case where no document contains the answer, the system says so instead of guessing.
Where your time actually goes
Translating product needs into AI solutions. Deciding whether to use LLMs, RAG, agents, or simple rules.
Building pipelines that feed the model with the right context: prompt engineering, retrieval systems, embeddings.
Working with models and iterating on outputs. Testing prompts, system instructions, and comparing responses for quality.
Engineer
Designing evaluation for AI: hallucination, safety, consistency. Adding guardrails and optimizing latency and cost.
Deploying AI systems and continuously improving based on real usage, failures, and prompt drift.
The hardest part isn't the model, it's controlling it. Getting an LLM to work once is easy. Getting it to behave reliably, avoid hallucinations, respect constraints, and not break in production, that's the real job.
Part three
Where the skills overlap and where they diverge
All three roles share a foundation. The divergence happens fast once you go beyond it. The column for your selected career is highlighted.
| Dimension | Data Scientist | ML Engineer | AI Engineer |
|---|---|---|---|
| Core language | Python, SQL | Python | Python |
| Trains models? | Sometimes, classical ML | Yes, that is the role | Rarely, uses APIs |
| Primary output | Insight and decision | Model and production system | Working AI powered product |
| Key concepts | Statistical analysis, data visualization, experiment design, storytelling with data | Supervised and unsupervised learning, model evaluation, training pipelines, optimization | RAG, agents, prompt engineering, LLM APIs, evaluation |
| Infrastructure | Minimal, notebooks and BI tools | Heavy, cloud and deployment pipelines | Medium, APIs and cloud services |
| Math depth | Statistics, probability | Linear algebra, calculus, optimization | Conceptual understanding |
| Closest analogy | The detective | The factory engineer | The product builder |
What has become essential across all three roles in 2026: production mindset, GenAI literacy, cloud deployment basics, and the ability to communicate technical work to non-technical stakeholders.
Part four
The reality of the market, especially in Lebanon
The clean separation between these three roles exists in mature tech companies with large teams. In most Lebanese companies, and honestly most companies regionally, the reality looks very different.
One title, many jobs
In a startup, you will likely be hired under one title and asked to do three jobs. You clean the data, build the model, deploy it, build the dashboard, and then get asked to wrap it all in an internal tool too.
This is not unique to AI. A backend developer becomes DevOps becomes frontend when the team is small enough.
Why this happens
Many Lebanese companies do not yet have a clear understanding of what these roles actually are. They know they need someone who "does AI" and figure out the details later.
Small teams cannot afford specialization. One person covering multiple roles is not a failure, it is how early-stage companies survive.
Knowing where you sit in the bigger picture, even if your job does not reflect it yet, gives you clarity on what to learn next and where you want to grow. The roadmaps ahead are built around that clarity.
Part five
The question that cuts through the confusion
If you are standing at this crossroads, ask yourself where in the pipeline you want to live. Click each card to reveal your path.
Scientist
Roadmap coming soon
Engineer
Roadmap coming soon
Engineer
Roadmap coming soon
None of these is more technical than the others. They are technically deep in different directions. Choose based on which problems energise you, not which title is trending.