Case Study - Speak your website into existence

Digital Hearth builds a website by letting you talk. Practitioners, healers, and nature guides answer 15 spoken questions, and the AI writes a published page and their brand language in about 20 minutes.

Client
Digital Hearth
Year
Service
AI product design, strategy, full-stack development

Overview

A somatic practitioner, an herbalist, a breathwork facilitator. Their gift is unmistakable in a room. Put them in front of a blank text field and the words never match the work. Digital Hearth closes that gap by letting them talk instead of write.

The practitioner answers fifteen questions across three acts: Your Work, A Real Story, What Makes You Different. Most people choose to speak, because the things worth saying about this kind of work tend to arrive out loud before they land on the page. As they talk, the questions pick up a thread from earlier and follow it, and what comes back sounds like them at their most clear, not like a brochure.

Out of that conversation comes a published practitioner page and nine pieces of brand copy, a one-liner, an about page, service descriptions, a transformation story, and social posts, all in the practitioner's own words. They can edit any of it, adjust the tone with AI suggestions, download a PDF, and publish to a custom subdomain.

We built Digital Hearth as the self-service version of our one-on-one brand coaching, so a practitioner can arrive with their voice and leave with a page that sounds like them.

How we solved it

The trap with generating brand copy is that you get nine pieces that each sound fine and nothing that sounds like one person. A one-liner, an about page, and a set of social posts have to carry the same voice or the whole thing reads as written by a committee. So the engine holds on to the practitioner's actual words from the conversation and reaches back to them when it writes each piece, rather than working from a tidy summary that has already sanded the character off. Voice input in a browser is the other soft spot, since speech recognition is uneven across devices, so we built the flow to fall back to typing at any point without losing the thread the practitioner was on.

What we built

  • AI product design & strategy
  • Voice-first conversational UI
  • Claude AI engine
  • Next.js & MongoDB
  • Web Speech API
Questions to a published page
15
From voice to website
~20 min
Brand copy pieces generated
9
Primary input method
Voice

More case studies

A field guide for taking the class outside

Rewyld for Schools is a teacher's field guide for leading weekly outdoor mindfulness with a class. Prompt-based, not step-by-step, following the actual seasons, with zero teacher prep. A Rewyld × Plantenvironment collaboration, now in pilot with London primary schools.

Read more

Ground truth for AI agents acting in the physical world

Pointmoon is ground truth for AI agents acting in the physical world. It gives a model a sourced, time-stamped read of the present moment (weather, season, water, land, sky), or an honest silence when the truth will not hold. It hands over the fact and lets the model do the talking.

Read more

Let's work together