30 days of machine learning: a project-based plan (free Notion template)
Skip the 40-hour theory course. This 30-day machine learning plan ships three real projects: statistical analysis, a supervised model with proper validation, and feature engineering with XGBoost, tracked in a free Notion template.
The classic way to learn machine learning is to spend two months on theory, burn out around gradient descent, and never train a model on messy real data. The project-based way inverts it: ship a small end-to-end project, hit a wall, learn exactly the theory that unblocks you, repeat. This 30-day plan is the second way.
This is the Data Science track of our free 30 Days of Data Notion template. Fair warning: it is the most demanding of the four tracks, closer to 75-90 minutes a day than 30. If that cadence sounds wrong for your month, the template's Foundations mode runs the same plan at two projects instead of three. Curious what the job itself looks like? Read a day in the life of a data scientist.
What 30 days gets you (and what it does not)
Thirty days will not cover neural networks, MLOps or LLMs. It will make you someone who frames a problem, prepares data, trains and validates models properly, and communicates results, which is the actual daily job, and the part most self-taught candidates skip straight past on their way to transformers.
The 30-day plan
- Days 1-3
Python and statistics intro
The free intro course: Python working knowledge and the statistics you will actually use. Solo alternative: Kaggle Learn's Python and Intro to ML micro-courses.
- Days 4-10
Statistical analysis with hypothesis tests
Real exploratory analysis on real-world data with tested, defensible conclusions. Solo alternative: the World Happiness Report dataset. The unglamorous project that teaches you to distrust your first chart.
- Days 11-18
Supervised learning, done properly
A full scikit-learn workflow: train-test discipline, cross-validation, metric choice, error analysis. Solo alternative: the Titanic competition, which is a cliché precisely because it is the right size.
- Day 19
Portfolio v1 goes live
Two projects online with write-ups a non-ML reader can follow. Communicating results is half the job; the write-up is not optional homework.
- Days 22-26
Feature engineering and XGBoost
Where tabular ML gets real: engineered features, gradient boosting, and understanding why it wins. Solo alternative: the House Prices competition.
- Days 27-30
Polish, publish, share
Recruiter test on day 28, publish on day 29 with #30DaysOfData, retro on day 30.
Guided or solo
Every project offers two routes: guided on D8A Academy with the dataset, a stepped brief, GitHub validation and an auto-built portfolio page, or solo with the Kaggle competitions named above. For ML specifically, the guided route's biggest gift is honest evaluation criteria: it is very easy to believe a leaking model is working when nobody defines "done" for you.
Day 0 takes ten minutes. Pick DS, pick your mode, block your daily slot, and make today day 0.
