Hard truth first: recruiters don't read certificates, they skim proof of work. Five solid projects beat fifteen course badges, every single time. Here's exactly which 5 to build, in order, with data that costs โน0.
๐งฑ The rules before you start
1๏ธโฃ SQL deep-dive โ the interview skill
Take a real transactional dataset (retail orders, food delivery, Olist Brazilian e-commerce โ all free on Kaggle). Load it into PostgreSQL or SQLite and answer 10 business questions with pure SQL: top repeat customers, monthly retention cohorts, revenue by category over time.
Proves: you can query โ the #1 tested skill in analyst interviews. Put the 10 questions + queries in the README.
2๏ธโฃ The messy-data cleanup
Grab a genuinely dirty public dataset (government open data portals are gold for this โ missing values, mixed formats, spelling chaos). Document every cleaning decision in a notebook: what you dropped, what you fixed, and why. Our free datasets field guide shows where to find these and how to clean them step by step with pandas.
Proves: you survive real-world data โ 80% of the actual job.
3๏ธโฃ The dashboard a manager would actually open
Pick one domain (sales, IPL stats, your own expenses โ anything) and build ONE clean dashboard in Power BI (free) or Looker Studio (free). Rules: max 5 charts, one clear headline insight at the top, filters that work.
Proves: you can communicate, not just calculate. Screenshot it into the README for the non-clickers.
4๏ธโฃ Exploratory analysis with a point of view
Take a dataset about something you genuinely like โ movies, cricket, stocks, gaming โ and do a full EDA that ends with 3 surprising findings written in plain English. Not "correlation is 0.7" but "sequels earn 40% less on average, except in horror."
Proves: curiosity + storytelling. This is the project that gets remembered in interviews.
5๏ธโฃ The end-to-end capstone
One project that chains it all: raw data โ clean โ SQL/analysis โ dashboard โ one-page written summary with a recommendation ("if I were the business, I'd do X"). Host the summary as a blog-style post and link everything.
Proves: you can own the whole pipeline โ this is the "hire them" signal.
๐ซ Mistakes that get portfolios binned
๐ The 4-week plan
TL;DR โ 5 projects, free data, public repos, READMEs with real findings: SQL โ cleanup โ dashboard โ story โ capstone. That's the whole formula.
Working on a portfolio right now? Drop your project idea below โ happy to tell you which of the 5 slots it fills best. ๐
๐งฑ The rules before you start
- Every project gets a public GitHub repo with a README that explains the question, the data, and what you found โ the README is the portfolio, the code is the appendix
- Every project answers a business question, not a tutorial prompt. "Which products drive repeat orders?" hits. "Titanic survival prediction" does not.
- One project = one weekend, two max. Shipped beats perfect.
1๏ธโฃ SQL deep-dive โ the interview skill
Take a real transactional dataset (retail orders, food delivery, Olist Brazilian e-commerce โ all free on Kaggle). Load it into PostgreSQL or SQLite and answer 10 business questions with pure SQL: top repeat customers, monthly retention cohorts, revenue by category over time.
Proves: you can query โ the #1 tested skill in analyst interviews. Put the 10 questions + queries in the README.
2๏ธโฃ The messy-data cleanup
Grab a genuinely dirty public dataset (government open data portals are gold for this โ missing values, mixed formats, spelling chaos). Document every cleaning decision in a notebook: what you dropped, what you fixed, and why. Our free datasets field guide shows where to find these and how to clean them step by step with pandas.
Proves: you survive real-world data โ 80% of the actual job.
3๏ธโฃ The dashboard a manager would actually open
Pick one domain (sales, IPL stats, your own expenses โ anything) and build ONE clean dashboard in Power BI (free) or Looker Studio (free). Rules: max 5 charts, one clear headline insight at the top, filters that work.
Proves: you can communicate, not just calculate. Screenshot it into the README for the non-clickers.
4๏ธโฃ Exploratory analysis with a point of view
Take a dataset about something you genuinely like โ movies, cricket, stocks, gaming โ and do a full EDA that ends with 3 surprising findings written in plain English. Not "correlation is 0.7" but "sequels earn 40% less on average, except in horror."
Proves: curiosity + storytelling. This is the project that gets remembered in interviews.
5๏ธโฃ The end-to-end capstone
One project that chains it all: raw data โ clean โ SQL/analysis โ dashboard โ one-page written summary with a recommendation ("if I were the business, I'd do X"). Host the summary as a blog-style post and link everything.
Proves: you can own the whole pipeline โ this is the "hire them" signal.
๐ซ Mistakes that get portfolios binned
- Titanic / Iris / MNIST clones โ recruiters have seen ten thousand; it signals "followed a tutorial"
- Code with no README, or a README with no findings
- Ten half-finished repos instead of five finished ones
- Charts with no takeaway sentence under them โ always write the "so what"
๐ The 4-week plan
- Week 1 โ project 2 (cleanup) + project 1 (SQL)
- Week 2 โ project 3 (dashboard)
- Week 3 โ project 4 (EDA story) โ post the 3 findings on LinkedIn too, recruiters lurk there
- Week 4 โ capstone + polish every README like it's a product page
TL;DR โ 5 projects, free data, public repos, READMEs with real findings: SQL โ cleanup โ dashboard โ story โ capstone. That's the whole formula.
Working on a portfolio right now? Drop your project idea below โ happy to tell you which of the 5 slots it fills best. ๐
๐ Like0โค๏ธ Love0๐ Haha0๐ฅ Fire0๐ Clap0โก 0