Ubuncare

The problem

Mental health support can be difficult to access consistently, while generic AI chat experiences often lack continuity, privacy clarity and appropriate boundaries.

Why we did it

We wanted to explore whether conversational AI could provide a more private, accessible and continuous wellbeing experience, without pretending to replace professional care.

Our approach

We treated memory and safety as product and trust problems, not just technical features. The experience was designed around user control, contextual continuity, a clear separation between live conversation, summaries and long-term memory, and clear boundaries around what the product is and is not for.

The solution

Ubuncare is a voice-and-text AI mental health and wellbeing app built around ongoing emotional check-ins, contextual memory, post-conversation summaries, user-controlled memory, privacy controls and support and safety signposting. It is designed as a space for reflection and continuity between conversations. It does not replace a therapist, doctor or emergency mental health service.

How we went about it

We explored the product through rapid prototyping, mobile-first UX iteration, realtime voice architecture, memory-model design, privacy principles, safety and boundary considerations, and repeated testing of conversational controls and user flows.

The decisions that mattered

• Raw audio was not used as the source of memory; conversations fed post-conversation summaries rather than being stored as permanent recordings.
• Live conversation, summaries and long-term memory were kept as separate layers, rather than treating every message as part of one continuous permanent record.
• Persistence required user approval, so information only became part of long-term memory after the user allowed it, rather than by default.
• Voice interaction was treated as a distinct design problem from text chat, with its own controls rather than reusing the text experience.
• The product was deliberately positioned as wellbeing support rather than clinical or emergency care, with support and safety signposting built into the experience rather than left implicit.

Outcome

A working prototype demonstrating realtime voice interaction, contextual memory, user-controlled persistence and privacy-aware design for a mental health and wellbeing context.

What we learned

Persistent memory is as much a trust and safety problem as a technical one in a mental health context. The features that mattered most were the ones giving people visibility and control over what was remembered, and clarity about what the product is not, as much as what it is.