Most roadmaps show you everything there is. This one shows you a route: one line everyone rides, then one line you choose, and a pile of things you can safely ignore for now. Every station is a few lessons on this site, and every line ends in something you build, not a list of topics.
4 lines to choose from 25 stations about 11 hours of lessons on the shortest route
Swipe the map sideways to see every line. Pick a station to see its lessons. Stations fill in as you finish lessons on this site, and a ring marks where you left off. Nothing is sent anywhere: it’s read from this browser.
Build the server side of apps: APIs, databases, logins.
Pick it if you like making things other people use, and you want to see them working in a browser.
Leads to Backend developer · Python web developer · API developer
Core + this line: about 14 hours of lessons.
Turn messy tables into answers people act on.
Pick it if you like finding the pattern in a spreadsheet more than building the spreadsheet.
Leads to Data analyst · Business intelligence analyst · Analytics engineer (later)
Core + this line: about 13 hours of lessons.
Models that predict, and apps built on language models.
Pick it if you want to build things that learn from examples, and you're happy to do the Data line's first stations first.
Leads to Junior ML engineer · Data scientist · AI application developer
Core + this line: about 18 hours of lessons. Rides the Data line as far as Clean & explore, then changes.
Scripts that do the boring parts of any job.
Pick it if you have a job already, or want one, where an hour of Python saves a day of clicking.
Leads to Automation / QA engineer · IT or operations roles · Any job, done faster
Rarely a job title on its own; it's the line that makes every other job easier, and the fastest way to use Python at work this month.
Core + this line: about 11 hours of lessons.
The first five Core stations are the same whichever line you pick. By When it breaks you’ll know whether you enjoy building tools, finding patterns or getting rid of chores, and that is your answer. All three big lines lead somewhere real: in the latest Python Developers Survey, 51% use Python for data analysis, 46% for web development and 41% for machine learning.
Everyone rides this one. It ends when you can build a small tool, test it, and keep it in Git.
Values, names and the REPL. Run every example; reading code isn't the same as writing it.
Most beginner programs are text in, decision, text out. This station is that loop.
Lists and dictionaries carry almost every program you'll ever write. Get fluent here.
You’re ready to move on when you can build a word-frequency counter for any text file.
Start from: Word Frequency Counter →The point where scripts become programs: name a step, reuse it, import it, and run it from a terminal on your own machine.
Reading a traceback calmly is the skill that separates people who finish from people who stall.
You’re ready to move on when you can build a to-do list that survives being closed and reopened.
Start from: TODO List CLI (with Persistence) →Classes when they earn their place, paths that work on every OS, an environment per project.
You’re ready to move on when you can build a file organiser that tidies a messy downloads folder.
Start from: File Organizer (Tidy a Messy Folder) →Git, tests, logs and config. Nobody lists them as exciting; every employer checks for them.
You’re ready to move on when you can build a command-line tool with tests, in a Git repo with a README; add the tests to the CLI calculator.
Start from: CLI Calculator (with History) →Build the server side of apps: APIs, databases, logins.
Be a client before you're a server: once you've called an API, building one makes sense.
FastAPI first: the most-used Python web framework in the 2024 survey, and type hints you'll reuse everywhere.
Prefer Django? Swap this station for Django: Batteries-Included Web, Django Models & ORM, Django REST Framework.
PostgreSQL is the default answer in production; learn the patterns that keep an ORM fast.
Hash passwords, issue tokens, and walk the checklist before anyone else walks it for you.
A project only counts once it runs somewhere other than your laptop.
You’ve arrived when you can build a deployed API: the URL shortener rebuilt with FastAPI, with logins, tests that run on every push, and a README that says why.
Start from: URL Shortener →Turn messy tables into answers people act on.
What a dataset is, where it comes from, and how to load your own instead of a tutorial's.
pandas is the most-used data tool in Python by a distance. Everything after this station builds on it.
Real data is 80% cleaning. Analysts who are good at it are the ones teams keep asking for.
A finding nobody understands didn't happen. One honest chart beats ten pretty ones.
Most company data lives in a database, not a CSV. Every analyst job asks for SQL.
You’ve arrived when you can build an analysis of a real dataset: the question, the cleaning, three charts and what you'd tell a manager.
Start from: the practice datasets →Models that predict, and apps built on language models.
Predicting a number, predicting a category, and measuring whether you did. No maths degree required.
Trees and forests win on most real tables. Pipelines stop your test data leaking into training.
Build one from scratch once, so Keras stops being magic.
Where most new AI jobs actually are: prompting, retrieval, and knowing when a model is guessing.
A model in a notebook is a draft. Put it behind an API and say honestly how often it's wrong.
You’ve arrived when you can build a model behind an API, with an evaluation that admits where it fails.
Start from: Deploy Your Neural Network →Scripts that do the boring parts of any job.
Rename, sort, search and report: the scripts that pay for your learning in week one.
Pull data from APIs and web pages, push it into spreadsheets and files.
A script you have to remember to run isn't automation yet. Schedule it, log it, keep its secrets out of the code.
You’ve arrived when you can build a scheduled job that does real work every day and tells you when it fails; the weather CLI, run each morning by GitHub Actions, is a good first one.
Start from: Weather CLI →Every one of these turns up in someone’s “must learn” list. None of them is what stands between you and a first job. Some are on this site; they’ll still be here when you need them.
They explain how Python itself works, not how to build things with it.
When it’s time: When you're writing a framework, or a library that other people's classes plug into.
It’s here when you need it →They make simple bugs hard to find, and most beginner programs are fast enough already.
When it’s time: When a profiler shows you're waiting on the network or the CPU. On the Web line, async arrives with FastAPI.
It’s here when you need it →Django, Flask and FastAPI solve the same problems. Knowing one well beats knowing three badly.
When it’s time: When a job you want uses the other one; the ideas carry straight over.
It’s here when you need it →They solve the problems of a big team running many services. You have one app and no team.
When it’s time: When you're in that team. Docker, on the Web line, is the part worth knowing now.
Patterns are names for solutions you'll meet anyway; memorised first, they turn into over-built code.
When it’s time: After a couple of projects, when you recognise the problems they name.
It’s here when you need it →Interesting, and almost never what a junior ML job asks you to build.
When it’s time: When you're past the ML line and want depth for its own sake.
It’s here when you need it →Useful for some interviews, but a hundred puzzles teach less than one finished project.
When it’s time: A few weeks before interviews that you know use them. Easy and medium are plenty to start.
They change monthly. The ideas underneath (prompts, retrieval, evaluation) don't.
When it’s time: When one solves a problem you actually have.
It’s here when you need it →Not a certificate. These are what a hiring manager can check in the two minutes they give your application. Tick them off as you go; the ticks stay in this browser.
About the hours: they count reading time, and reading is the small part. Building the checkpoint projects takes longer than every lesson on your route put together, and it’s where the learning sticks.