Building Your Data Science Portfolio

Why a Portfolio Matters

A resume tells someone what courses you took and what tools you list. A portfolio shows them what you can actually do with those tools — a working project they can open, read, and evaluate for themselves. For data science specifically, this matters because the skill being hired for (turning messy data into a working analysis or model) is exactly the kind of thing that's hard to verify from a bullet point but easy to verify from a real project.

You don't need dozens of projects. You need a small number of projects that are genuinely finished, genuinely yours, and genuinely explained — not copied tutorial code with the variable names changed.

Anatomy of a Great GitHub Project

GitHub is the default place to host a data science project — it's free, it's what nearly every recruiter or hiring manager already knows how to browse, and it doubles as your version-control history. A well-organized repository looks roughly like this:

student-performance-predictor/
├── README.md              # the front door — read this first
├── requirements.txt        # exact packages needed to run it
├── data/
│   └── student_performance.csv
├── notebooks/
│   └── 01_eda_and_model.ipynb   # the full walkthrough, in order
├── src/
│   └── train_model.py      # the same logic, as a runnable script
└── images/
    └── confusion_matrix.png    # a saved chart for the README

The single highest-leverage file in that structure is README.md — it's the first (and often only) thing a visitor reads. A strong README answers four questions, in order:

# Student Performance Predictor

Predicts whether a student will pass or fail an upcoming exam
from three easy-to-collect numbers: weekly study hours,
attendance percentage, and past exam average.

## Data
`student_performance.csv` — 2,000 rows, 4 columns. This is a
synthetic dataset built to practice the full ML workflow, not
a real published study.

## Approach
1. Clean missing attendance values (median fill)
2. Explore correlations between features and the pass/fail label
3. Stratified 80/20 train/test split
4. Train a RandomForestClassifier (200 trees, max_depth=6)
5. Evaluate with classification_report + a confusion matrix

## Result
80% test accuracy. `past_exam_avg` was the single strongest
predictor (feature importance 0.47), ahead of study hours (0.31)
and attendance (0.22).

## How to run
    pip install -r requirements.txt
    python src/train_model.py

Writing a Short Project Summary or Blog Post

A README is written for someone who's already looking at your code. A project summary — a short post on a personal blog, a LinkedIn article, or even just a pinned post — is written for someone deciding whether to look at your code at all. It's shorter, less technical, and leads with the interesting part.

LinkedIn Profile Tips for Data Science Roles

LinkedIn is where a lot of the informal discovery in data science happens — recruiters search it, and other data scientists share work on it. A few concrete, low-effort habits go a long way.

Turning This Course's Two Projects Into Portfolio Pieces

You already built two complete, end-to-end projects in this course. Neither is portfolio-ready as-is — but both are close. Here's exactly what each one needs.

📊 Project 2 — Sales Dashboard (Section 3)

This project already demonstrates real skills: Matplotlib/Seaborn charting and a Plotly Dash interactive dashboard. To make it portfolio-ready:

🎓 Lesson 30 — Student Performance Predictor

This is your strongest ML project — a full pipeline from raw CSV to an evaluated, interpreted model. To make it portfolio-ready:

Lesson Summary