About us

Good speakers aren't born.
They practise.

navanmai exists because access to great speaking coaching has always been unequal. A one-on-one speaking coach costs hundreds of dollars an hour. Most people never get that feedback — so they keep presenting the same way, with the same habits, wondering why they don't feel confident.

Our mission

We want to make high-quality, objective speech coaching available to anyone with a microphone. navanmai gives you the kind of feedback a professional coach would give — what pace you're speaking at, where you're pausing, how your pitch moves, how many filler words you're using — and then helps you practise until those habits change.

What we built

Record yourself speaking in your browser — freeform on any topic, or reading from a prepared script. navanmai transcribes your recording, extracts detailed audio features, and then synthesises them into personalised coaching feedback.

The app includes a pronunciation coach for drilling individual words, a delivery challenge mode that scores your performance 1–10 against randomised business scenarios, and a session history so you can track your improvement over time.

Who it's for

navanmai is built for anyone who wants to speak more clearly and confidently. That includes professionals preparing for presentations or pitches, students practising for debates or academic talks, job seekers honing their interview delivery, and anyone who's ever wished they could hear themselves the way their audience does.

How the technology works

Your recording is transcribed locally on our server using OpenAI Whisper — your raw audio never leaves our infrastructure. We then extract objective metrics (speaking rate, pitch contour, energy levels, pause detection, filler word counts) using audio processing libraries. These structured metrics are passed to the Claude AI to generate readable, actionable coaching feedback.

Separating the analysis from the feedback keeps the process deterministic and cost-efficient: the AI is synthesising measurements, not guessing from raw audio.

Get in touch