USAAIO Prepunofficial
USA AI Olympiad, 2027 season

A free online course for USAAIO Round 1 and Round 2

Short lessons, Part 1 style questions with worked solutions, coding problems graded like the contest, and timed mock exams. It runs in your browser, with no account.

168lessons
858Part 1 style questions
114coding problems with graders
812syntax cards
1043glossary terms
15Round 1 units

Who it is for

Strong math students preparing for USAAIO Round 1 and Round 2, and for IOAI after that. It assumes you are comfortable with algebra and functions. It teaches the rest from the start: Python, NumPy, pandas, PyTorch, the linear algebra, calculus and probability that machine learning uses, classical machine learning, neural networks and CNNs, and how to work under exam conditions.

What is inside

Learn and practice

  • 168 lessons in 15 Round 1 units, 10 to 15 minutes each, with interactive figures and code you can edit and run.
  • 858 Part 1 style questions (five choices, or an integer 000 to 999) with worked solutions and the reason each wrong choice is wrong.
  • 812 syntax cards: you write the line of NumPy, pandas or PyTorch from memory and it is checked by running it.
  • A glossary of 1043 terms, and summary sheets to print.

Test yourself

  • 114 coding problems with hidden tests and the contest's import and function bans.
  • Unit quizzes, progress checks every two weeks with a projected Part 1 score, and full timed mocks.
  • A Today page that plans each session from the minutes you have, with spaced review of what you missed.
  • 8 Round 2 units (transformers, generative models, reinforcement learning, IOAI tools) follow Round 1.

How it works

Python in your browser

Code runs in Pyodide, Python compiled to WebAssembly, with NumPy, pandas, scikit-learn and Matplotlib. It downloads once (about 30 MB) the first time you press Run, and is cached after that.

Colab for PyTorch

PyTorch does not run in a browser. PyTorch problems open in Google Colab, where the real exam runs, with the starter code and a grader cell. Every problem has a Colab notebook, so you can also time your NumPy solutions on Colab's computers.

Your progress stays with you

Answers, times and scores are saved in this browser and nowhere else. You can export them to a file and import them on another computer.

Spaced review

Questions and cards you miss come back after 1, 3, 7, 16 and 35 days, so the review goes to what you have not mastered yet.

The 2027 contest in brief

StageWhenWhat
Round 1Friday, February 19, 2027Part 1: 20 questions in 60 minutes, multiple choice or an integer 000 to 999, no resources. Right 7.5, wrong 0, blank 1.5 points.
Part 2: 75 minutes of coding in Google Colab, CPU only, with two organizer documents as the only resource.
Round 2March or April 2027By invitation. The whole syllabus, including transformers and generative models.
CampJune 2027Team USA is chosen at camp.
IOAIJuly 2027The International Olympiad in Artificial Intelligence.

From the published 2027 information. The official rules are on usaaio.org; check there before the contest.

Where to start

Already know some of it?

The placement quiz has 20 questions across the Round 1 units (about 40 minutes). It tells you, unit by unit, what to skip, skim or learn, and the planner then skips the units you test out of.

Take the placement quiz

Starting fresh?

Lesson 1 explains how the contest works, how Part 1 is scored (it decides when to guess), and how to use this site. Then the Today page plans each session.

Start with lesson 1

Questions

Is this the official USAAIO site?
No. It is an independent study course, not affiliated with or endorsed by USAAIO. For rules, dates and registration, use usaaio.org.
Does it cost anything, or need an account?
No and no. There are no accounts, ads or tracking.
Where is my progress saved?
In this browser only. Clearing site data erases it, so export a copy now and then (the About page links to export and import). An export can be imported on another computer.
Which questions are official?
None of the official problems or samples are included. Questions are original or adapted from past contests and labelled as such.
Does the code really run?
NumPy, pandas, scikit-learn and Matplotlib run in your browser with Pyodide (Python compiled to WebAssembly), downloaded the first time you run code. PyTorch does not run in a browser: lesson examples that use it show output recorded when the site was built, and PyTorch problems open in Google Colab with a grader cell.
Why is there a Colab link on NumPy problems?
Browser Python is slower than Colab and its timing varies, so time limits here are approximate. Each problem's Colab notebook has the same grader with the real limits.
Can I use it on a phone?
Lessons, questions, syntax cards, quizzes and progress work on a phone. The code lab works too, but it is much easier on a laptop.
Does it work offline?
After your first visit to the course, yes: the lessons and questions are cached. Python needs one online run to download, and is cached after that.
I found a mistake.
Every lesson, question, card and problem has a "Report a mistake" link. It opens a prefilled GitHub issue for you to finish and submit; nothing is sent automatically.