Build a Kaggle Portfolio for Data Science Jobs and Freelancing

In an interview, someone asked me: "You say you know Python. Show me."

My resume said "Python, Pandas, some Machine Learning." What I actually had to show were two half-finished college assignments that weren't even on GitHub. That interview is where it hit me: saying you know something and being able to show it are two completely different things.

Six months after that interview, I have 6 published notebooks on Kaggle, and my first paying client came from one of them.

---

What Kaggle actually is, if you've never used it

Kaggle is a site where people practice data science in public. There are two main things on it: datasets (free, uploaded by anyone, anything from IPL scores to restaurant reviews to startup funding data) and competitions (companies post a real problem, people submit models, and the best one wins a prize).

But the part that actually helps a career is notebooks. It's a Jupyter-style coding environment where you take a dataset, analyze it, chart it, maybe train a model, and publish the whole thing in one place. People can read your code, run it themselves, and upvote it.

That published notebook becomes your free, public portfolio.

---

My first notebook, and why it was bad

My first notebook used a cricket dataset, IPL match scores. I figured "cricket data, this'll be interesting." Threw it together. A few charts, a conclusion that said "Mumbai Indians is the best team" (which, I later realized, wasn't even statistically true).

Zero upvotes. Zero comments. I assumed nobody actually looks at Kaggle notebooks.

Actually, that's not quite right. People were looking. My notebook just wasn't worth stopping for.

The real problem was different: my notebook was just code. No explanation anywhere. Nobody reading it could tell what I was doing or why I was doing it.

I went and looked at the top notebooks on other datasets. Almost every code cell had a line above it: "Here I'm removing outliers because..." or "This chart shows that..." They were narrating their thinking, not just dumping code.

---

The second notebook is where things changed

I picked a dataset that genuinely interested me, restaurant reviews from Bangalore. This time I did things differently.

I wrote a short note above every section explaining what I was looking for and why. Where the model didn't perform well, I said so directly: "This approach isn't working well because there are too many missing values in the ratings column." I wrote the honest limitation instead of hiding it.

I also added one small chart showing which cuisines got searched more on weekends. A small insight, but a specific one.

That notebook got 34 upvotes, 3 comments, and one person who started following me on LinkedIn. Small numbers, sure. But compared to the first notebook, it was a different universe.

---

Where the first client actually came from

My third notebook used a small e-commerce dataset, predicting customer churn. A startup founder commented on it: "Can you do something similar for my dataset? DM me."

My first thought was that it was spam. I replied anyway.

Turned out to be genuine. A small direct-to-consumer company wanted to know which customers weren't coming back to buy again. I took their anonymized data, ran a similar analysis to what I'd shown in the Kaggle notebook, and charged Rs 3,340 for it.

First money I ever made from data science. Not from an interview, not from a resume. From a public notebook someone happened to read.

---

The mistake to avoid: don't just copy tutorials

A lot of people pick the "Titanic Survival Prediction" dataset, the one every YouTube tutorial uses, and follow the exact same steps shown in the video. Same code, same charts, sometimes even the same variable names.

That impresses nobody. A recruiter spots a copied notebook almost instantly.

What works better: pick a dataset with some local or unusual context that hasn't already been analyzed a thousand times the same way. Regional sports data, delivery times for a food app, city-level census numbers, anything that has your own angle on it rather than a repeated tutorial script.

---

What an actual good notebook needs

A solid notebook usually follows a similar shape.

A short intro explaining what you're trying to find and where the dataset came from. Then data cleaning, where you explain how you handled missing values or odd entries instead of silently dropping them. A couple of visualizations in the middle that actually say something, not decoration for its own sake. And an honest summary at the end covering what worked and what genuinely didn't.

Length doesn't matter much. My most successful notebook was about 40 lines. Another one was closer to 200. Both worked because both were clear about what they were doing.

---

What to actually do if you're starting from zero

Make a free Kaggle account today. Find a dataset that genuinely interests you, from any industry, something you'd actually enjoy digging into. Your first notebook probably won't be great. Mine wasn't either.

By the second or third one, the pattern starts making sense. Build a GitHub portfolio alongside it too, and link both to your LinkedIn profile. The process for starting data analyst freelancing on Fiverr follows roughly the same shape. Only the source of the dataset changes.

A client doesn't show up on the day you start. Mine took three notebooks. For some people it's faster. For others, it might take ten.