Living legacy vaultsFamily legacy, easier to revisit
A family can ask about consented, recorded moments and receive answers rooted in real experiences — clearly labelled when the archive has no answer.
Yovaan AI
Memories you can talk to.
Yovaan AI transforms a person's own conversations — with their consent — into a Digital Being: their voice, their traits, their way of speaking. Able to speak when they no longer can.
Working software PoC · Research preview
The problem
We save the photos. We rarely save the way someone spoke.
Pictures and videos help us remember how a person appeared. But the story of the individual — their expressions, their advice, their way of talking to the people they loved — is something no file format holds on to.
Static media can't converse
Photographs and videos capture what someone looked like. None of that data captures how they thought, joked, or what they knew best.
People speak differently with each other
One person jokes with a sibling, mentors a student, and speaks formally at work. Those distinctions deserve to be preserved — not averaged into a generic profile.
Everyday talk feels more natural
Formal interviews are valuable, but informal conversation — with permission — carries the phrases, stories, and rhythms people actually use.
Stories fade within a family
Without a way to organize and revisit them, family stories, wisdom, and sayings become harder and harder to pass on.
Digital Being · Concept
Our vision
Remember more of a person — with their permission.
Our long-term vision is to convert consented conversations into a digitized, interactive version of a person — software first, with a wearable as a future design direction. A Digital Being that belongs to the person it represents, governed by their consent.
01
Consent comes first
A Digital Being is trained and recorded only with the person's explicit permission — revocable, auditable, and deletable at any time.
02
Learn the relationship, not just the voice
Everyone speaks differently to family, friends, and colleagues. Personas stay separated per relationship, never averaged into one generic voice.
03
Show what the evidence can support
Responses are grounded in recorded moments with source references — and the system abstains honestly when the archive has no answer.
04
A staged, measurable presence
Fidelity grows in validated stages, from text sketch to voice-true presence. We measure what can be represented faithfully — never promise a full replica blindly.
Coin-sized wearable · Design direction
Software first. Hardware only when the evidence supports it.
Progressive fidelity
A digital representation should show its limits as it grows.
We explore a staged model that makes evidence coverage visible. More recording time does not automatically mean more fidelity — every stage is evaluated against held-out conversations.
00–20%
Early profile
Text chat with basic vocabulary and facts. Recognizably them on paper.
20–50%
Style signals
Their slang and catchphrases appear. Relationships stay separated.
50–80%
Evidence-grounded
Tone and cadence emerge. Grounded recall of shared moments, with sources.
80–100%
Long-term goal
Voice synthesis approaches their real voice. Near-exact manner of speaking.
Long-term goal: measure what can be represented faithfully at each stage — rather than promise a "full replica."
Applications
If the core works safely, several futures open up.
Different interfaces, same underlying data. Every application below depends on the same foundation: consent, evidence, and honest disclosure.
Living legacy vaults
Elder care · With cliniciansA possible assistive-care use case
A known, consented archive could support reminiscence and connection for elders. Clinical applications would always be designed with specialists.
Knowledge transferPreserve how people teach
A future research path: organize a person's explanations, mentoring style, and craft so the next generation can keep learning from them.
Creators · OTT · Rights-firstLicensed interactive characters
Under the rights-holder's approval, authors could explore interactive works with sanctioned characters and scripts — a new canvas for storytelling.
How impactful can it be
From raw audio to a living presence.
One system, one source of truth: the person's own words. Every interface — chat, voice, or a future wearable — draws from the same consented, evidence-grounded memory.
Pipeline: raw audio becomes speaker-labelled transcripts, persona profiles, a searchable memory index, evidence-grounded responses, and finally an interactive Digital Being.
Privacy · Security · Consent
Built in at every stage — not bolted on.
A Digital Being is only trustworthy if the person it represents is in control. These principles are structural, not features.
The system does not do voice cloning today. It always identifies itself as AI, abstains when uncertain, and never impersonates without consent.
Explicit consent
From all speakers, before anything is recorded, processed, or learned.
Access control & encryption
User-level isolation, encrypted at rest and in transit.
Consent audit logs
Every action tracked. The person can always see what happened and why.
Right to revoke & delete
Consent is understandable, specific, easy to withdraw — and deletion is real.