How to Make a Data Scientist Portfolio

The standard data science portfolio is a GitHub profile containing four Jupyter notebooks, one of which is the Titanic dataset.

The problem is not that notebooks are bad. It is that a notebook is a record of your working, and a portfolio needs to be a record of your conclusions. Nobody reviewing candidates is going to read 300 cells to find out whether you are any good.

---

Lead With the Finding, Not the Method

Every project should open with what you found, in plain language:

Analysed three years of municipal bus GPS data and found that 60% of delays on one route came from a single intersection.

Then the method, then the code. A reader who stops after the first line still knows whether you can do the job. A reader who stops after your import statements knows nothing.

---

Choose Projects With a Real Question

The dataset matters less than the question. Three characteristics of a project worth including:

A question someone would actually ask. Not "explore this dataset" but "which factors predict churn in this cohort, and how confident can we be?"

Data you had to work for. Scraping, joining several sources, dealing with genuinely messy input. Kaggle datasets arrive pre-cleaned, which removes most of the real work.

A conclusion you can defend. Including the limitations. Saying "this model is 94% accurate but the classes are imbalanced and accuracy is the wrong metric here" demonstrates more than a higher number would.

Projects to Avoid

Titanic survival. MNIST digits. Iris classification. Boston housing. These appear in thousands of portfolios, and using them signals that you have completed a tutorial rather than solved a problem.

---

Communicate Like the Audience Is Not Technical

Much of data science is explaining findings to people who will not read your code. Your portfolio is where you prove you can do that.

For each project include:

• One or two clear charts. Labelled axes, a title that states the finding, no chartjunk.
• A short written summary. Three or four sentences, no jargon.
• The technical detail, below, for people who want it.
• A link to the repo for people who want all of it.

A chart whose title is "Figure 3" is a wasted opportunity. "Delays cluster between 4pm and 6pm on weekdays" does the work for the reader.

---

Show Your Reasoning About Limitations

This is the single most underused signal in data science portfolios.

State what your analysis cannot support. Where the data was thin. What confounders you could not rule out. What you would collect next.

Junior candidates present results. Stronger candidates present results with appropriate uncertainty — and hiring managers can tell the difference immediately.

Structure and presentation examples are in these developer portfolio examples, which apply closely to technical portfolios generally.

---

Frequently Asked Questions

How many projects should a data science portfolio have?

Three or four, properly written up. One excellent end-to-end project beats six notebooks.

Do I need deployed models?

Not necessarily, but at least one project that goes from raw data to something usable — a dashboard, an API, a report someone could act on — separates you from candidates who stop at the model.

Is a GitHub profile enough?

No. GitHub shows code to people who can read code. A portfolio shows judgment to everyone else, including the recruiter doing the first screen.

Should I include academic work?

Yes, if you can state the question and finding without assuming domain knowledge. Rewrite the abstract for a general reader.

---

Build It Free with Folify

You do not need to hand-code a site or pay a monthly fee to get a portfolio online. Folify is a free portfolio builder with a drag-and-drop block editor, custom domains, and built-in analytics on the free plan.

Pick a starting point from the portfolio templates, drop in your work, and publish. Most people are live in under ten minutes.

Start building free →

---

No credit card. No coding. Just your work, looking its best.