What a data scientist
portfolio should include.
A data scientist portfolio should lead with three to five projects, each stating the problem, the data, the method, the model, and the measured result, such as an AUC, an RMSE, a precision and recall figure, or the lift measured in an A/B test. Every project needs a linked, runnable GitHub repository with a clear README, a stated environment and random seed, and a data source that is public or synthetic, because proprietary employer data and any personally identifiable information in a dataset must never appear on a public page. Below is the full list of what to put in, the terms a recruiter's applicant tracking system searches, and which of the Portfolio designs suit a technical body of work.
The sections a data scientist portfolio needs.
A data scientist is judged on reproducible technical work, not a job title, so the portfolio is organised around proof you can frame a question, build a model, and show what it produced. Work through these in order, and read the flagged block twice.
Three to five flagship projects
Pick a small set over a long list. For each one, state the problem you set out to answer, where the data came from, the method and model you chose and why, and the measured result: an AUC, an RMSE, a precision and recall figure, or the lift a test produced. A reviewer decides on the strongest three, not the busiest twelve.
Linked, runnable code
Each project links to a real GitHub repository with a README that explains the setup, not a screenshot of a notebook. State the environment, pin the dependencies, fix the random seed, and name the data source, so someone else can run it and get the same number you reported.
A dashboard or chart for a non-technical reader
At least one project should include a chart or a short dashboard that communicates the result to someone who will never open the notebook. A hiring manager often reads the chart before the code, so it has to carry the finding on its own.
Your stack, stated plainly
Name the tools rather than implying them: Python, SQL, pandas, NumPy, scikit-learn, PyTorch or TensorFlow, notebooks, Git, and the cloud platform if you deployed anything. A recruiter's search and an applicant tracking system both key off these exact words.
End-to-end thinking, not just a model score
Show how you framed the question, engineered the features, validated the model against a holdout or cross-validation, and what happened after training, whether that was a deployed endpoint, a batch job, or a handoff to an engineering team. A score with no context around it is hard to trust.
Competitions and open-source work
A Kaggle ranking, a published write-up, or a merged pull request to an open-source library are extra proof a reviewer did not have to take on faith. List them briefly, they carry more weight alongside your flagship projects than on their own.
Never include: proprietary data, PII, or NDA-covered numbers
No dataset, screenshot, or metric that belongs to a past employer, and no personally identifiable information in any sample data you show, even scrubbed. If a project used company data, rebuild the idea on a public dataset such as Kaggle, UCI, or a government source, or generate synthetic data, and say so plainly in the README.
The same applies to a specific number covered by an NDA. Describe the shape of the work and the type of result in general terms instead of publishing the figure itself.
Terms a data science recruiter searches.
Recruiters and applicant tracking systems search for specific tools and techniques. If these are true of you, use the exact words, because a system indexes the words you wrote, not the ones you meant.
Paste your resume into the free ATS score checker with a real data science job posting to see which of these terms the posting uses and your resume is missing.
Which designs suit a technical body of work.
Data science hiring reads projects, not just a title, so the design has to show a grid of work and get out of the way of code and charts. Of the 60 Portfolio designs and 48 resume layouts, these are the shapes that fit.
Pick a design built around a grid of project cards, each opening to a page with the problem, the method, the result, and links to the repo, the notebook, or a live dashboard. Avoid a single long-scroll design meant for one big case study, it does not scale to five projects.
Of the 48 resume layouts, choose a single-column one rather than a two-column design. Multi-column resumes can serialise into a scrambled reading order when a system parses them, which undercuts a resume built to show a clean chain of tools and results.
Order the page so your strongest project and its measured result sit near the top, ahead of a personal statement. A reviewer wants evidence of a shipped model or a validated result before they read why you like the field.
Use a design with a monospaced font available for code snippets and a chart style that stays legible at a glance. A reviewer scanning between projects should be able to read a result without squinting at a screenshot.
Who a data scientist portfolio is not for.
A portfolio helps some data scientists and is invisible to others. Read this before you spend a weekend on one, because the payoff depends heavily on how you actually get hired.
Worth building if you
- +Are switching into data science from another field and need to show working code, not just a job title.
- +Have a handful of projects you are proud of and want one link that shows the problem, the model, and the result for each.
- +Are applying to a startup or a smaller team that reads GitHub and a portfolio site alongside a resume.
- +Do Kaggle competitions, publish write-ups, or contribute to open source and want those in one place.
Skip it, for now, if you
- −Only apply through a referral where your resume gets passed directly to a hiring manager, a site rarely gets opened in that path.
- −Have work that is entirely under NDA with nothing public to show. Build one public project on open data first, then come back.
- −Are targeting a pure research role hired on a publication record. That reviewer is checking Google Scholar and a paper list, not a portfolio site.
- −Have limited time before a deadline. Fix your resume for the ATS first, then build the site.
Questions data scientists ask.
Straight answers on projects, proprietary data, and whether the effort is worth it.
Can I show results from my current job on my portfolio?
Not the company's data or the exact figures if they are proprietary or covered by an NDA. Rebuild the idea on a public dataset, describe the approach at a general level, and say plainly in the README that the project uses public or synthetic data standing in for a real problem you solved at work.
How many projects should I put on my portfolio?
Three to five, chosen for range and depth rather than volume. One classification or regression problem with a clear metric, one project that shows an end-to-end pipeline including deployment, and one that leans on a different skill, such as NLP or an A/B test, cover more ground than a long list of half-finished notebooks.
Do I need a live dashboard or is a notebook enough?
A notebook proves the work happened. A dashboard or a chart built for a non-technical reader proves you can communicate the result to the person who will actually decide whether to use it. Include both where you can, they answer different questions for a reviewer.
Do data scientists even need a portfolio to get hired?
Often not for a role filled entirely through a referral or an internal transfer. A portfolio earns its keep when a resume alone cannot show that a model was actually built and validated, which is most external applications, career switches, and Kaggle or open-source driven job searches.
What if all my best work is under NDA?
Build one public project before you build the portfolio. Pick a public dataset close to the kind of problem you solve at work, run it through the same process you would use professionally, and write it up honestly as a demonstration rather than a real production result.
Where to go next.
Build the site, test your resume, or read how the paste-a-resume flow works.
Turn your data science
resume into a site.
Paste your resume and Portfolio drafts a clean, project-driven website in about a minute. A card grid for your work up top, no proprietary data or PII anywhere, published to your own domain with TLS handled for you.