Case Study - An outdoor mindfulness app for the restless mind

Overview

You want to feel present and awake, and sitting still on a cushion has never been the way in. Rewyld meets you on a walk instead. Step outside, press play, and a guide talks with you for about five minutes. Then the phone goes back in your pocket and you keep walking.

The practice is made for the walk you are actually on. It speaks to a wet November morning by the water, to the oak you are passing, to the cold on your face, because it is written the moment you press play from the real conditions around you. A warm, natural voice reads it to you.

Follow it day after day and the walks build into a journey. When you want to go deeper, the guide network connects you with certified practitioners from Kripalu, outdoor training programs, and partner organizations to walk with in person.

We designed and built the whole thing, from strategy and product design through launch. A Next.js web platform. A React Native app on iOS and Android with subscriptions through Apple and Stripe. A Node and Postgres backend. Claude writes each practice against live place data from Pointmoon, so the words match the ground you are standing on. ElevenLabs voices the script, and we stream it chunk by chunk so it starts fast and plays without gaps. Guides list their practices at guides.rewyld.earth, and place pages at rewyld.earth bring new walkers in from search.

How we solved it

Writing and voicing a practice at press-play takes tens of seconds, and no one wants to stand still watching a spinner. We hid the wait inside the experience. The opening breath screen holds for about twelve seconds while generation runs behind it, and by the time it finishes the practice is usually ready. We also cut the audio into paragraph chunks and synthesize them at once, so the first chunk starts playing in seconds while the rest arrive and join on without a gap.

Two smaller problems shaped the craft. The voice model reads a written pause shorter than it feels on a walk, so we stretch every pause in the script before it reaches the voice, since that is where a practice earns its calm. And a generated voice drifts in loudness from one take to the next, so we master every practice to one level and it plays at the same volume every time.

A practice also has to be honest about the place. Rather than let the model guess the weather or the season, we ground every generation in Pointmoon, our own place-truth service. That closed off a class of bugs where a wrong coordinate would put a walker under the wrong sky.

What we built