What does a Data Scientist actually do?
Less "build the model," more "figure out which question is worth answering"
A data scientist spends less time on the glamorous parts of machine learning than people expect, and more time on data cleaning, exploration, and framing the right question in the first place — the classic "80% of the work is data prep" isn't far off in practice.
The role blends three skill sets: enough statistics to know whether a result is real or noise, enough programming to wrangle data and build models at scale, and enough domain and communication skill to explain what it all means to people who don't read code.
It also overlaps heavily with neighboring roles. Data analysts lean more toward reporting and dashboards on existing data; ML/MLOps engineers lean more toward productionizing and operating models at scale. Data scientists usually sit in between — exploring, modeling, and prototyping before something graduates into production.
The Data Scientist Roadmap
Work through these in order, from statistics to a model someone else can act on
Data Scientist Fundamentals
What the role actually covers, and how it differs from adjacent titles.
Programming Fundamentals
The languages that show up in nearly every data science job description.
Mathematics & Statistics
The theory that separates a real finding from a coincidence in the data.
Data Wrangling & Cleaning
Where most of the actual working hours go — getting messy data into usable shape.
Exploratory Data Analysis
Understand what's actually in the data before building anything on top of it.
Databases & Big Data
Query, store, and process data at a scale a single laptop can't handle.
Machine Learning Fundamentals
The core algorithms and how to honestly evaluate whether they're working.
Advanced Machine Learning
Where accuracy usually comes from once the basics are solid.
Model Deployment & MLOps Basics
Get a model out of a notebook and in front of something that can call it.
Data Storytelling & Communication
An insight that never reaches a decision-maker didn't really happen.
Tools & Ecosystem
The everyday tooling around a working data scientist's laptop.
GitHub Projects
Real, buildable projects to put on your own GitHub
Data Science Notebooks
Work through a large collection of notebooks covering Pandas, scikit-learn, and deep learning to see idiomatic patterns across the stack.
Kaggle Competition Notebook
Enter a beginner-friendly Kaggle competition and write up your full EDA-to-model pipeline as a public notebook for your portfolio.
Build a Streamlit Dashboard
Turn an analysis into an interactive Streamlit app so a non-technical stakeholder can explore the data themselves.
Made With ML
Follow a complete project from data exploration through model training and a deployed API to practise the full lifecycle end to end.
Frequently Asked Questions
Common questions from people starting out in data science
Data Scientist vs Data Analyst vs ML Engineer — what's the difference?
Data analysts mostly report on and visualize existing data to answer defined business questions. Data scientists go further — framing open-ended questions, building predictive models, and running experiments. ML/MLOps engineers focus more on productionizing and operating those models at scale.
Do I need a master's or PhD to become a data scientist?
It's not required for most industry roles — a strong portfolio of real projects and solid fundamentals in statistics and programming can substitute for a formal degree, though some research-heavy roles (particularly in R&D or academia-adjacent fields) still expect one.
Should I learn Python or R?
Python is the more common default in industry today, with a broader ecosystem for production and deployment. R still holds strong in academia, biostatistics, and anywhere statistical rigor and visualization are the primary deliverable. Many data scientists end up comfortable with both.
How much math do I actually need?
Enough statistics to interpret a result correctly and enough linear algebra and calculus to understand how the models you're using actually work — you don't need to derive gradient descent from scratch, but you should know what it's doing and why it can fail.
What should go in a data science portfolio?
A small number of complete, well-documented projects beats a large number of shallow ones — ideally covering the full pipeline from messy raw data through EDA, modeling, and a clear write-up of what was found and why it matters.
Is data science still in demand with AI advancing so fast?
Yes, though the day-to-day work keeps shifting — less time on routine model-building that off-the-shelf tools now handle, more time on data quality, experiment design, and knowing which questions are actually worth answering with data.
Track complete
From statistics and data wrangling to EDA, machine learning, and communicating results — that's the core of what's expected of a working data scientist. Keep building, and put it to work on a real dataset.
Where next?
Keep exploring by domain or drill into a single skill