How CodeKelas works
Every course runs one loop: understand the code, find what is wrong, improve the spec, and go round again.
- 1
Understand
Read the AI's code before you run it, and say what it will do.
How it is checked
Questions about your own code, marked by running it, and a spoken explanation every week.
- 2
Find
Prove what is wrong with a test that fails, or show with tests that nothing is.
How it is checked
Your tests run against broken versions of the code to see what they catch.
- 3
Improve the spec
Fix the instructions the AI was given instead of patching what it wrote.
How it is checked
The AI writes the code again from your spec, several times; a good spec gives working code every time.
- 4
Go round again
Run the checks again and answer every review comment before you merge.
How it is checked
Work counts once its checks pass and every comment has an answer.
Back to Understand, round again until the code holds
The one-to-one, minute by minute
Sixty minutes, in the same order every week. Your tutor prepares from a record of your week in the app, so no time goes on catching up.
- 00:00ExplainYou explain part of your week's pull request aloud, without AI.
- 10:00ReviewYour tutor works through the comments, your replies and any disagreements.
- 25:00The new ideaA worked example, then you predict and trace before you try it.
- 45:00Without AIA short task to trace, fix or extend. Once a month this is the unassisted check.
- 55:00The planNext week's work, the next Git step you take over, and one skill to practise.
When AI is allowed, and how much
Each piece of work carries one rule. The app enforces it for the AI it provides; your answers and your spoken explanations show whether you understood the work.
- L0No AIWhat AI may do
Nothing.
What you doTrace, write and explain unassisted.
Where it is usedMonthly checks, vivas, the weekly explanation, timed exam practice
- L1HintsWhat AI may do
A nudge, then the idea, then the line at fault, then a worked example of a similar problem.
What you doTry first. The next hint opens only after another attempt.
Where it is usedThe basics and exam practice in every course
- L2ExplainWhat AI may do
Explanations of code, errors and ideas, and feedback on your code. No solution code.
What you doWrite the code yourself, and note what the explanation changed.
Where it is usedDebugging and test-writing exercises
- L3WriteWhat AI may do
Code, once your spec and tests are in.
What you doPredict what it does before you run it, then explain every line you keep.
Where it is usedMost building exercises
- L4AgentWhat AI may do
An agent that plans and edits several files on a branch.
What you doBrief it with a spec, review its diff, answer the questions, write the pull request description.
Where it is usedCampus from Month 1, Build with AI from Week 4, Work, and supervised Teen capstones for ages 16 and 17
A separate permission lets a student's own program call an AI model. Every such call goes through CodeKelas and is logged, and personal data never goes into one.
How your work is reviewed
Every pull request goes through four reviews before it counts, much as a junior engineer's would at work.
- 1
Automatic checks
Within minutes: the hidden tests, and your own tests run against broken versions of the code.
- 2
Questions about your change
You answer questions about the change you made. Your answers are marked by running your code.
- 3
AI review comments
Short comments that point at a problem and ask a question, but never write the fix. For students under 18, a blocking or security comment waits for the tutor first.
- 4
Your tutor
Your tutor reviews every milestone, anything flagged and a sample of the rest before your next one-to-one.
You answer every comment, by changing the code or by giving a reason, and you can dispute an AI comment you think is wrong. A correct dispute earns credit.
How the certificate mark is made
Half the mark comes from work done without AI or spoken aloud; on Work it is 40%. The certificate lists the hidden-test suites your code passed.
- Checks without AI in the one-to-one25%
- Weekly explanations and the final viva25%
- Exercises and questions, marked automatically15%
- Projects and milestones, marked by your tutor25%
- Answering review, disputes included10%
Without AI, or spoken aloud50%
On GitHub from the first week
Every piece of work is a real commit and pull request in your own repository. At first the app does each Git step for you and shows you what it did; then you take the steps over one at a time.
GitHub requires users to be 13 or older. Repositories stay private while a student is under 18.
- 01Make the repository
- 02Save a version (a commit)
- 03Work on a branch
- 04Commit and push
- 05Open the pull request
- 06Undo a change
- 07Review someone else's pull request
- 08Resolve a merge conflict
If you have never coded
- Nothing to install for the first two weeks: the editor runs in the browser, on a laptop or a Chromebook.
- No blank page in the first lesson. You read a short working program, predict what it does, run it, and change one line.
- Reading before writing: predicting output, tracing step by step, putting given lines in the right order.
- From the second week, short 'find what is wrong' tasks on code an AI wrote.
- Help offered after a failed run, without waiting to be asked, and a soft deadline the day before each one-to-one.
Ask about a place
Tell us who the course is for and which one you are looking at. We will reply with the dates and fees.