How to Build a Portfolio for Remote Data Science Jobs

That matters even more from Nigeria. A hiring manager in Nairobi, London or Toronto may get hundreds of applications for one remote role. They decide in a minute or two whether your GitHub link is worth opening. This guide explains which projects to pick, how to structure each one, where to host them for free (as of October 2026) and how to match your portfolio to a real job ad. It also covers working around power cuts, data costs and the wrong kind of datasets.

Quick answer

Build three to five focused projects, each tied to a business question an employer would care about: forecasting, churn, fraud, pricing or a dashboard. Give each project a README that states the problem, the data, the method, the result and how to rerun it. Host the code on GitHub, add one or two live demos (Streamlit Community Cloud is free as of October 2026) and link everything from your CV and LinkedIn.

What a remote data science portfolio has to prove

An office-based employer can judge you over weeks of sitting near you. A remote employer has your CV, your portfolio and a few video calls. Your portfolio therefore has to answer questions that would normally be answered in person.

• Can you frame a problem? The best projects start with a business question ("Which customers are likely to stop paying in the next 60 days?") rather than a library ("I used XGBoost").
• Can you work with messy data? Cleaning, joining and checking data takes much of the real job. Show it.
• Can you explain results to non-technical people? Remote teams make many decisions in writing. A README or short report that a product manager can follow is evidence of that skill.
• Can someone else run your work? Pinned package versions, a clear folder structure and instructions show you can hand work over across time zones.
• Do you use judgement? Say what didn't work, what the model gets wrong and what you'd do next. Honest limits read as seniority.

If your portfolio shows all five, it does a job your CV can't.

Data science, analytics or machine learning: decide which jobs you're targeting

"Data scientist" covers very different jobs, and one portfolio can't serve all of them. Pick a lane before you build anything.

If you're still deciding between these paths, our data scientist career guide and data analyst career guide cover the skills and routes in. If you're moving from analysis towards modelling, moving from data analysis to machine learning publishes with this series.

How many projects, and which kinds

Three to five finished projects beat twelve half-finished notebooks. Recruiters skim, so depth in a few projects is more convincing than breadth.

A balanced set for most remote data science roles looks like this:

You don't need every row. A BI-leaning candidate might do two SQL projects and two dashboards. An ML-leaning candidate might do one analysis, two models and one deployed app.

Projects to leave out or rework

The Titanic, Iris and MNIST datasets are teaching exercises. Hiring managers have seen thousands of them, and they prove you followed a tutorial. The same applies to copying a Kaggle notebook and changing a few lines. If a tutorial project taught you something, keep it in a "learning" folder and don't pin it.

Unfinished projects also hurt you. A repository with a notebook called Untitled3.ipynb and no README tells a recruiter you don't finish things.

Choosing project ideas that don't look like homework

The strongest projects answer a question a real business has, using data you had to find, clean or combine yourself. Three ways to get there:

1. Start from job ads. Open five remote data science ads you'd apply for and list the problems they mention (fraud, credit scoring, pricing, retention, marketing attribution). Build towards those.
2. Use local knowledge as an advantage. Nigerian and African markets are under-represented in portfolios. A project on mobile money transaction patterns (using synthetic or public data), food price movements from National Bureau of Statistics releases, or naira exchange-rate volatility stands out from yet another house-price model. For exchange rates, the CBN exchange rate page publishes official figures.
3. Combine sources. Joining two datasets (say, weather and agricultural prices, or exchange rates and import data) shows data-wrangling skill that single clean CSV files never will.

Example (hypothetical): Tolu is targeting fintech roles. Instead of a generic credit-risk notebook, she builds a synthetic dataset modelled on how buy-now-pay-later customers behave. She trains a default model, explains which features matter and why, and writes up what a lender should do differently. She then deploys a small app where a user enters a profile and sees a risk score with an explanation. The project lines up with the problems in fintech job ads, so it gives her something concrete to talk about in interviews.

A note on data privacy and confidentiality

Never publish personal data about real people: names, phone numbers, BVNs or account details, even if you found them online. Use public, anonymised or synthetic data. Don't publish anything from a current or past employer's systems either, even "just the code". If you want to show work you did on the job, rebuild the idea with public or synthetic data and describe it in general terms.

How to structure each project

Every project should be readable in two minutes by someone who will never run your code, and runnable in fifteen minutes by someone who will.

Repository layout

A simple, predictable structure helps reviewers:

• README.md: the project's front page (more below)
• data/: a small sample or a script that downloads the data, not a 2 GB file
• notebooks/: exploration, numbered in order (01_explore.ipynb, 02_model.ipynb)
• src/: reusable code (cleaning functions, training scripts)
• requirements.txt or environment.yml: pinned package versions
• app/: the demo app, if there is one

What the README must say

Write the README for a hiring manager first and an engineer second. Use these headings:

1. Problem: one or two sentences in business language.
2. Result: the headline finding or metric, near the top, with what it means in practice.
3. Data: where it came from, its size, its time period and its known problems.
4. Approach: the steps you took and why you chose them over the alternatives.
5. Limitations: what the model gets wrong, any bias in the data, what you'd need to deploy it for real.
6. How to run: exact commands.
7. Demo: link to the live app or dashboard.

Example (hypothetical) result line: "The model flags 70% of customers who later defaulted while reviewing only the riskiest 20% of applications. A lender could use it to prioritise manual checks, not to approve or reject automatically."

A line like that shows business judgement. "Achieved 0.91 AUC" on its own doesn't.

Make metrics honest

Use a proper train and test split (or time-based validation for anything involving dates), and compare against a simple baseline. If your fancy model beats "predict the average" by only a little, say so. Interviewers will ask, and an honest README makes that question easy to answer.

Where to host your data science portfolio

You need three layers: code on GitHub, live demos for one or two projects, and a single landing page that links everything. All three can be free.

GitHub and GitHub Pages

GitHub is where most technical reviewers will look first. Pin your three to five best repositories to your profile and add a profile README that lists them with one-line descriptions.

For a landing page, GitHub Pages lets you publish one user site per account from a repository named after your username followed by .github.io (for example, yourusername.github.io), plus one project site per repository. GitHub's published limits cap a site at 1 GB with a soft bandwidth cap of 100 GB a month, which is plenty for a portfolio of text, charts and links. Don't host datasets there.

Streamlit Community Cloud

Streamlit Community Cloud lets you create, deploy and manage Streamlit apps for free by connecting your GitHub repositories. That makes it the easiest way to turn a Python model into something a recruiter can click and use. Keep the app light: load a pre-trained model file rather than training on start-up, and use a small sample of data.

Hugging Face Spaces

Hugging Face Spaces suits machine learning demos, but read the current terms before you plan around it. According to Hugging Face's Spaces documentation (checked October 2026), static HTML Spaces are free for everyone. Gradio and Docker Spaces need a paid plan to create, although free personal accounts can still host up to two Gradio Spaces running on ZeroGPU. Spaces on free hardware also go to sleep after a period without use, so a recruiter may wait for yours to wake up. Mention that in your link text, or prefer Streamlit for your main demo.

A personal website: optional

A custom domain looks polished, but it's not required. A clean GitHub profile plus a simple GitHub Pages site does the same job. Spend the time on a better project instead.

Read the job ad first, then shape the portfolio

Your portfolio should mirror the tools and problems in the ads you're applying to, and those ads also tell you if you can apply at all.

As an example, ALX Africa's Data Scientist ad on Greenhouse (open when checked on 27 October 2026) is full-time, remote within Africa, with preferred time zones of SAST, CAT and EAT (GMT+2 to GMT+3). Lagos is on WAT (UTC+1), so a Nigeria-based hire would be one to two hours behind the team. The ad asks for:

• 3 to 5 years of hands-on data science, analytics or BI experience
• strong SQL with complex queries and large datasets
• Python for analysis and automation
• dashboards in QuickSight, Metabase or Power BI
• clear communication with non-technical stakeholders

It doesn't publish a salary range. Instead, it asks for your expected gross annual salary in USD, and the application asks for a CV and a short essay of up to 200 words, not a portfolio link. That's common: many ads don't ask for a portfolio directly. Recruiters still open the GitHub or portfolio link on your CV to check claims such as "strong SQL" and "dashboards in Power BI". A candidate targeting this kind of role should lead with a SQL-heavy analysis and a Metabase or Power BI dashboard, not a deep learning project. Check the live ad before applying, because eligibility and requirements change.

Before you build, sort target ads by where applicants may live:

• Worldwide or "anywhere": open to Nigeria-based applicants unless the ad says otherwise.
• Africa, EMEA or named countries: check that Nigeria is included and that you can work the stated time zone.
• US-only, UK-only or EU-only: usually requires existing work authorisation or residence there, so a better portfolio won't change the answer.

Listings on WorkRemote's remote tech jobs show where applicants may live, which saves you this step.

Building a portfolio from Nigeria: power, data and cost

Most portfolio work can be done on a mid-range laptop with no paid tools. The bigger risks are lost work and wasted data, and both are easy to manage.

• Commit often. Push to GitHub at the end of every session. A power cut or a dead laptop then costs you an hour, not a project.
• Plan around power. Run long training jobs when you have stable power from the grid, an inverter or solar. Save model checkpoints so a crash doesn't send you back to the start.
• Keep datasets small. Work with samples of a few hundred megabytes or less. Large downloads eat mobile data, and reviewers won't download them anyway.
• Use free cloud notebooks and hosting sparingly. Train locally or in a free notebook environment, then deploy only the small trained model.
• Have a backup connection. A MiFi plus a phone hotspot means a failed upload doesn't wait until tomorrow.
• Record a short walkthrough video. If an app sleeps or fails during a recruiter's visit, a two-minute screen recording linked in the README still shows it working.

Putting your portfolio in front of recruiters

A great portfolio that nobody opens does nothing. Put links where reviewers already look.

1. CV header: one GitHub or portfolio link next to your email and LinkedIn. Use a short URL, not a long string.
2. Experience or projects section: two or three projects, each with a one-line business result and a direct link.
3. LinkedIn Featured section: add your best project and your landing page. Our guide to writing a LinkedIn profile for remote job opportunities publishes with this series.
4. Cover note or application answers: link the one project closest to the role's main problem.
5. Interviews: have a project ready to screen-share and talk through in five minutes. For likely questions, see our data scientist interview questions.

Many remote data roles also include a take-home task. A portfolio built around reproducible, well-documented projects makes those easier, because you've already practised the format. Our guide to remote technical assessments publishes with this series.

How much the portfolio is worth: pay context

A portfolio doesn't set your salary, but it affects which roles you can reach. For a benchmark, the US Bureau of Labor Statistics puts the median annual pay for data scientists at $120,230 in May 2025, with employment projected to grow 35% from 2025 to 2035. That figure is US local pay for people working in the US. Remote offers to Nigeria-based contractors are usually well below it, and they vary widely by employer, seniority and whether the company pays location-based rates. Use the BLS figure as a ceiling for context, not as your asking price. For analyst-level pay in dollars and naira, see our remote data analyst salary guide.

Common mistakes that weaken a data science portfolio

• Leading with tools instead of problems. "Built a random forest" says less than "cut false fraud alerts by a third in a test set".
• No README, or a README that only says how to install. Reviewers won't open your notebook to find out what you did.
• Notebooks with errors or no outputs saved. Rerun from top to bottom before you push.
• Leaked secrets. API keys in code get scraped quickly. Use environment variables or the host's secrets settings.
• Inflated claims. Saying a project was "used by a company" when it wasn't will come out in an interview.
• Mismatch with target roles. A deep learning-heavy portfolio sent to BI roles, or the reverse.
• Ignoring eligibility. Polishing a portfolio for ads you can't legally take wastes weeks.

Beware of anyone who offers to "build your portfolio" or guarantee a data job for an upfront fee. Our guide on spotting remote job scams in Nigeria covers the warning signs.

Frequently asked questions

How many projects should a data science portfolio have?

Three to five finished, well-documented projects is enough for most remote roles. Each should show a different skill, such as SQL analysis, a dashboard, a predictive model or a deployed app. Recruiters skim, so quality and clarity matter more than quantity.

Can I get a remote data science job with a portfolio but no degree?

Sometimes, but expect some ads to ask for one. The ALX Africa ad above asks for a degree "or equivalent" in a quantitative field, and the BLS lists a bachelor's degree as the typical entry-level education in the US. A strong portfolio helps you make the case for equivalence, especially with smaller companies and startups. The planned guide on remote AI jobs that do not require a PhD covers related routes.

Do remote employers actually look at GitHub?

Many technical reviewers check it, particularly at shortlisting or before a technical interview, even when the application form doesn't ask for it. Make your pinned repositories readable without running any code. A clear README and a live demo let a busy reviewer judge your work in two minutes.

Should I use Kaggle competitions in my portfolio?

Yes, if you did meaningful original work and explain what you learnt, such as feature ideas, validation choices and why your approach ranked where it did. A copied public notebook or a beginner competition everyone does adds little. Business-framed projects with data you collected or combined usually say more about you.

Where can I host a data science portfolio for free?

As of October 2026, GitHub (code), GitHub Pages (a landing page up to 1 GB) and Streamlit Community Cloud (interactive apps deployed from GitHub) are all free. Hugging Face offers free static Spaces, and free accounts can host up to two Gradio Spaces on ZeroGPU. Most other Gradio and Docker Spaces need a paid plan.

Can I include projects from my current job?

Only if your employer has agreed in writing and nothing confidential is exposed, which is rare. The safer approach is to rebuild the idea with public or synthetic data and describe the original work in general terms on your CV.

How long does it take to build a portfolio?

That depends on your skills and time. A focused project with a clean README and a demo often takes a few weekends rather than a few days. Building one project well, getting feedback, then building the next is faster overall than starting five at once.

Key takeaways

• A remote data science portfolio has to prove in writing what an office employer would see in person: framing, cleaning, judgement and communication.
• Three to five business-framed projects, each with a clear README and honest metrics, beat a long list of tutorial notebooks.
• Match your portfolio to the ads you can apply for: their tools, their problems and their location rules.
• GitHub, GitHub Pages and Streamlit Community Cloud offer a free hosting setup as of October 2026. Check Hugging Face's current free-tier rules before relying on it.
• Use public, anonymised or synthetic data, and never publish an employer's confidential work.

Next steps

Pick the role you're targeting, open five live ads for it and list the tools and problems they share. Then build your first project around those. When you're ready to apply, browse remote jobs open to applicants in Nigeria and set up a free job alert for "data scientist" or "data analyst" so new matching roles reach your inbox daily. If you'd like a person on the WorkRemote team to rewrite your CV and LinkedIn around your portfolio, see what our CV review covers, which is part of Premium (₦29,999 a month). For the broader principles that apply across fields, our guide on building a portfolio that helps you get remote jobs publishes with this series.

Sources

• Data Scientists, Occupational Outlook Handbook, US Bureau of Labor Statistics
• Data Scientist, ALX Africa (Greenhouse job ad)
• What is GitHub Pages?, GitHub Docs
• GitHub Pages limits, GitHub Docs
• Streamlit Community Cloud, Streamlit Docs
• Spaces Overview, Hugging Face Hub documentation
• Exchange rates by currency, Central Bank of Nigeria