SpintoSpeak v2: A Partner That Talks Back

A year ago SpintoSpeak was one loop: spin a topic, talk for a minute, get scored. That loop worked, and you used it — thousands of times, in eight languages, at 6am before interviews and at midnight before demos. You also told us, repeatedly and specifically, what it was missing.

Almost everything in v2 came out of those messages. Thank you. Genuinely — this release is mostly your feedback with our name on it.

Here's what's new.

Two kinds of practice, two kinds of report

The old report scored every session the same way. But rehearsing "tell me about yourself" and riffing on a random prompt are different skills, and grading them identically helped neither.

v2 splits practice into two modes, each with its own report.

Topics

The open-ended mode, rebuilt around a single question: what's the one thing to fix next?

• One fix, up top. Not a list of twelve notes. The highest-leverage change, stated plainly, before anything else.
• Skill bars across relevance, confidence, clarity, pace, structure, and filler — so you can see the shape of your delivery at a glance.
• Structure, segment by segment. Your talk is split into opening, body, and closing, with how long you spent in each and what to do differently.
• Power words and weak words — the phrases that landed, and the ones that hedged.
• A polished version of what you actually said, so you can hear the gap between your draft and your best take.

Interview

A new mode built for answering questions under pressure.

• STAR breakdown. Situation, task, action, result — scored separately, because most weak answers are strong on situation and empty on result.
• Readiness score for the question you just answered.
• Five interview-specific scores: relevance, confidence, specificity, conciseness, and impact. Specificity is the one that surprises people.
• Hedging, counted. "Kind of", "I guess", "sort of" — every hedge, with examples pulled from your own transcript.
• Pauses, marked natural or awkward, plus your filler breakdown and words per minute.

Both reports still run on the same audio you already record. Nothing extra to set up.

An AI speaking partner — quiet, or in character

The biggest ask was some version of "I don't want to talk to a wall." v2 gives you three ways to run a session.

• Solo — the classic. Just you and the mic, no interruptions.
• Quiet support — you speak. The partner stays out of it and gives one short hint only when you actually stall. It won't open, won't monologue, won't take over your answer.
• Roleplay partner — the partner plays the other person. The interviewer, the skeptical customer, the friend you're arguing with. It opens in character, reacts, and asks real follow-ups.

Roleplay is the one worth trying first if you've only ever practiced alone. Being asked "wait, why did you choose that approach?" mid-answer is a very different exercise than talking into silence.

If you're practicing a language that isn't your native one, the partner never corrects you mid-conversation — that would wreck your flow. Corrections, accuracy, vocabulary range, and pronunciation all land in the written report afterward.

A rebuilt app

v2 is a full visual redesign, not a reskin. Cleaner surfaces, a real dark mode, a mobile layout that stops fighting your thumb, and a practice screen that gets out of the way while you're actually talking.

The spinner, the timer, and the report all got the same treatment: less chrome, more of the thing you came for.

Still eight languages

English, Türkçe, Español, Français, Português, 中文, العربية, and हिन्दी. Everything above works in all of them.

Start with one spin

If you haven't practiced in a while, don't plan a session. Open the app, spin, pick roleplay, and talk for one minute. The report will tell you the rest.

Take a spin →

And keep the feedback coming. It's clearly working.