You can now prove your skills by talking

Until this week there was one way to verify a project on Winumu: our AI read your evidence, wrote questions from it, and you typed your answers. That still works, and for a lot of people it is the right option. But typing has a ceiling. An answer written in a text box can be composed somewhere else, at any pace, with any amount of help.

A second, stronger way

There is now a live AI interview. It works like a video call. An AI interviewer that has already read your project sits on one side, your camera on the other. It asks about the real decisions inside your work, listens to your answer, and presses you when something is vague, the way a senior colleague would. You speak. There is nothing to type and nothing to paste.

It takes about eight minutes and asks three questions, or four if you are learning through an institution, plus whatever follow-ups your answers earn.

The questions come from your actual work

This is the part that matters. The interviewer is not working from a generic question bank. Before the call it reads your repository, your files, your description and your proof of output, and every question points at something concretely in there: a specific file, a function, a decision, a number.

Alongside each question it holds a private answer key built from the same evidence, which you never see. That is why a fluent, general answer about the technology does not pass, and why an awkward answer that names the right file does.

Your portfolio shows the difference

A project you defended out loud carries a gold "Live Interview Verified" badge instead of the standard green one, on your portfolio and on the page you send to employers. Speaking about your own work in real time, under follow-up questions, is harder to fake than a typed answer, and the badge says so.

Once you have earned it on a project it stays. Passing a written challenge on the same project later never takes it away.

What is recorded, and who sees it

The interview is recorded, and you are told before it starts. We keep your voice and a few still images from your camera, and the recording is checked against the transcript to confirm the conversation really happened. That check is precisely what makes this badge worth more than a typed test.

The recording stays private. Employers see your result. Institution staff can see the questions you were asked and what you answered. Neither sees the recording, and you can ask us to delete it at any time.

Built for real connections

A lot of our learners are on mobile data, on phones, on networks that drop. We built for that rather than around it:

• If your microphone, camera or connection will not hold, you are moved to a typed version of the same interview rather than locked out.
• If the call drops partway through, we save what was recorded and grade that instead of losing your attempt.
• The camera is optional. The interview runs on audio alone if you do not have one.
• Audio goes straight from your browser to the model, so the call uses as little of your data as we can manage.

On trying to game it

People will try, so it is worth being straight about what this does and does not do. The interviewer asks about things no amount of reading can tell you: what broke, what took longest, what confused you at the start. It follows up unpredictably, so pre-written answers come apart. We measure timing and compare the recording against the transcript.

What it is not is surveillance. We do not track your eyes or watch your screen, and we are not going to. Our aim is narrower and more honest: to make faking it cost more effort than learning the thing. If someone studies their own project deeply enough to hold up under real follow-up questions, they have done the work, and that is the whole point.

Trying it

Log or open a project, choose Verify, and pick the live AI interview instead of the written challenge. It is available on Pro. The written challenge is not going anywhere, and it still earns a real badge.