How it works
CodeSpeek is a coding-interview practice tool you use by talking. You describe the code out loud, an AI interviewer writes exactly what you said and asks a question when you are vague, your tests run in your own browser, and each attempt is graded into a review card that schedules the problem to come back.
1. Pick a problem
The statement, worked examples and a starter signature are on screen. Nothing is hidden behind a paywall or a timer.
2. Say what to write
“Loop over nums with enumerate, keep a hash map of value to index, and if target minus n is in the map return those two indices.” That is a complete instruction, and it becomes code.
Two things make this different from dictating into an editor. If what you said is ambiguous, you get asked instead of guessed at — “what should happen when nums is empty?” And if what you said is wrong, it is written wrong, so the failing test is yours to find.
3. Run the tests
Python runs in your browser through Pyodide. Your code is not uploaded to be executed anywhere, and a runaway loop is stopped after ten seconds instead of freezing the page.
4. Get nudged, not rescued
Ask for help and you get the lowest useful level first:
| Level | What you get |
|---|---|
| 1 | A question aimed at the gap — “what does your loop do on the second duplicate?” |
| 2 | A property of the problem, not the technique — “the input is sorted; does that buy you anything?” |
| 3 | The name of the pattern family |
| 4 | Pseudocode for one step |
| 5 | The approach — only after you say you want it |
That ladder is enforced by the server, not merely requested of the model: a reply that reaches for a higher level than you have unlocked is discarded before it reaches your screen.
5. Finish, and be scheduled
Finishing writes a review card — what you got, what you missed, how much help you took — and puts the problem back in the queue. A clean solve comes back in a week and then less often; a rescue comes back tomorrow.