VOICE / MOBILE AI
Haven AI
Keep the convenience of voice. Keep control of the journal.
A voice-first journaling experience with editable transcripts, deliberate AI assistance, and a device-local journal vault.
- My role
- Mobile & API development · Team contribution
- Status
- In development · Private source
- Built with
- React Native · Expo · TypeScript · SQLite
In development · Team project
The product idea
Speaking can be an easier starting point than a blank page. Turning a spoken thought into a useful journal entry still calls for care: transcription can be imperfect, generated wording may not sound like the person, and saving should be their decision. Haven AI puts review at the centre of that experience. It treats a recording, a transcript, an AI-generated draft, and a saved journal as distinct steps.
My contribution
Haven is a team project. My contribution spans the Expo mobile implementation, transcription integration, journal review and local storage, the supporting generation API, and development-build workflows. I worked on connecting the mobile experience to those capabilities while preserving clear boundaries between capturing a thought, asking for assistance, and saving an entry. The work also included Android transcription integration and build configuration for testing the native experience.
A deliberate voice-to-journal flow
A person starts a recording and reviews the resulting transcript. They can then choose a format such as a journal entry, notes, or a transcript-only record. AI assistance is a separate choice, and generated text remains editable before it is saved. Guided reflection provides another bounded way to work through a thought. The journal vault is device-local, keeping the saved writing within an experience the person can review and manage.
Status and learning
The product is in development; this showcase does not imply an app-store release. The images use a clean web preview without personal journal entries. Native recording and on-device transcription depend on a supported device and development build, so a web preview cannot demonstrate those capabilities on its own. The central lesson for me is that useful AI requires careful interaction design as well as model integration: the person needs to understand what happened and retain the ability to edit the result.
