Data Analytics Portfolio Projects: 3 That Get Interviews
A data analytics portfolio needs three projects: a decision dashboard, a SQL deep-dive, and a statistical notebook. Free public datasets and full specs inside.
Three finished projects are enough: a decision dashboard, a SQL deep-dive, and a statistical notebook. Together they answer the three questions every analytics interview circles back to. Can you turn data into a decision? Can you work a real database? Can you hold a number steady and say what it does and does not prove? Each project below comes with a free public dataset, a question worth answering, and the specific things a hiring manager looks for when opening it.
What the job asks you to provePermalink to “What the job asks you to prove”
The Occupational Outlook Handbook describes the work in plain terms: analyze data, use data visualization software to present findings, and make business recommendations to stakeholders. At the senior end of this track, data scientists earned a median of $120,230 in May 2025, against $50,980 for all occupations, and the Bureau of Labor Statistics projects employment to grow 35 percent from 2025 to 2035. The entry rung of the same track pays less; the data analyst salary page tracks what the role pays at each level.
A portfolio is your proof that you can do the listed work before anyone pays you for it. That is also why one finished portfolio project outweighs another certificate for most career changers; the tradeoff is laid out in certificates versus portfolios.
Project 1: A decision dashboard on real taxi dataPermalink to “Project 1: A decision dashboard on real taxi data”
New York's Taxi and Limousine Commission publishes monthly trip records in Parquet going back to 2009, with a taxi zone lookup table alongside. Since January 5, 2025, vehicles entering Manhattan's Congestion Relief Zone, the streets at or below 60th Street, pay a toll (MTA). From 2025 onward the TLC files even carry a cbd_congestion_fee column. That combination hands you a natural experiment for a dashboard: what happened to yellow-cab trips into the zone after pricing began?
Build it in Power BI Desktop or Tableau Public. Both are free, and the free analytics tools guide covers their exact limits. Pull six months of trips before the toll and six months after, join trips to zones, and put the answer on one screen.
What a hiring manager checks when opening it:
- The title states the question and the finding. "Taxi Dashboard" states nothing.
- One screen. A change-over-time view, a geographic view, and a summary number are enough.
- The numbers read like decisions: percent change against a baseline, not raw trip counts.
Project 2: A SQL deep-dive on a relational datasetPermalink to “Project 2: A SQL deep-dive on a relational dataset”
Spreadsheets end where joins begin, and interviews know it. IMDb publishes non-commercial datasets as gzipped tab-separated files, refreshed daily: title.basics, title.ratings, title.crew, name.basics. Load them into PostgreSQL, which developers have ranked the most desired and most admired database for three years running in the Stack Overflow survey, and you have a real relational model: titles, ratings, people, and credits that only make sense joined.
Pick a question with a shape, for example: have movies gotten longer by decade, and do better-rated ones run longer? Answering it takes joins across at least three tables, aggregation with care for vote counts, and window functions for within-decade ranks. Push the schema, the queries, and a short data dictionary to a free public repository.
What a hiring manager checks:
- Joins across tables, not one flat file filtered in Excel.
- SQL a colleague can read: named CTEs instead of a single 40-line query.
- A data dictionary, so a stranger can rerun the analysis without asking you anything.
Project 3: A statistical notebook that tests a claimPermalink to “Project 3: A statistical notebook that tests a claim”
The first two projects show you can deliver and query. The third shows judgment, the part where interviews go sideways. Our World in Data maintains a CO2 and greenhouse gas dataset published as CSV with a codebook that records a description and source for every indicator. That codebook is the professional habit on display: every column in your notebook traces back to a named source.
Take a question people argue about: do richer countries emit more CO2 per person? Load the CSV with pandas, plot on log axes, sit with the outliers instead of deleting them, and write down what the pattern supports and what it cannot. Correlation between income and emissions is a finding; causation is a claim you have not earned from a scatter plot.
What a hiring manager checks:
- The finding is stated in the first cell and repeated in the last, in plain language.
- Limits are named: what the data cannot tell you, and why.
- The notebook runs top to bottom on a clean machine.
The three projects at a glancePermalink to “The three projects at a glance”
| Project | Dataset | Free tools | The skill it proves |
|---|---|---|---|
| Decision dashboard | NYC TLC trip records | Power BI Desktop or Tableau Public | Turning data into a recommendation |
| SQL deep-dive | IMDb non-commercial datasets | PostgreSQL, DBeaver, GitHub | Working a real relational database |
| Statistical notebook | Our World in Data CO2 | Python, pandas, JupyterLab | Handling uncertainty honestly |
I can tell you what carried my own switch from sysadmin to analyst: a HealthTech dashboard built on the free stack. It was not the most technical thing I had ever produced. It answered one operational question from raw data through to a recommendation, and that completeness is what every interview kept coming back to.
Three mistakes that sink otherwise good portfoliosPermalink to “Three mistakes that sink otherwise good portfolios”
Tutorial datasets. Titanic, Iris, and the supermarket sample read as coursework, because that is where the reviewer last saw them. Public operational data reads like the job.
Dashboards without a question. Twelve charts and no verdict force the reviewer to do the analysis themselves. Name the question in the title and answer it in the first chart.
Notebooks that end mid-air. A notebook that stops after the last plot, with no conclusion, proves you can run code but not finish work. The last cell should repeat the finding in one sentence a non-analyst understands.
For the mechanics around the projects, hosting, ordering, and what to leave out, the complete portfolio guide for career changers is the hub. If you have no formal analytics history at all, portfolio projects with no experience shows how to borrow credibility from adjacent work.
How Traecta helpsPermalink to “How Traecta helps”
Traecta — Your Personalized Career Roadmap sequences these three projects against the skills you already have. If your background is heavy on spreadsheets, the dashboard milestone lands in your first month and the notebook gets extra weeks. If you already query databases at work, the SQL project moves first. Every milestone names its dataset and tool, so the portfolio assembles itself while you study.
What to take awayPermalink to “What to take away”
- Three finished projects beat a shelf of half-finished ones: a dashboard, a SQL analysis, and a notebook.
- Build on operational public data. Taxi records, IMDb, and CO2 series carry the messiness a reviewer wants to see you handle.
- State a finding in every project. Work without a verdict is decoration.
When you want the sequence scheduled around your background, your Traecta career roadmap puts the three milestones on a calendar and keeps the portfolio moving alongside the coursework.
