A safety-moderated AI companion that extends caregiver capacity
for persons living with dementia — without deception, without replacing the human.
55 million people live with dementia worldwide. Their caregivers — mostly family members — face unsustainable burdens with no safety net.
In dementia care, engagement is not a benefit. A vulnerable person will interact with any responsive system. The differentiator is constraint, oversight, and preservation of human relationships — not hours of AI airtime.
Carevatar deploys structured, caregiver-approved AI sessions — brief, bounded, and fully auditable. The caregiver stays in control. The person with dementia gets consistent, safe engagement between visits.
Native iOS/Android + web app. Set permissions, curate memory, review transcripts, receive alerts.
PLWD-facing tablet experience designed to feel like a familiar video call. No identity deception — ever.
Weekly longitudinal summaries: agitation trends, sleep signals, caregiver burden indicators — shared with PCPs.
Most AI companions are a single model with content filters bolted on. Carevatar is built differently: three specialized models working together.
Generates empathetic, context-aware conversation within caregiver-approved constraints. Creative but bounded.
Real-time pre/post screening of every message. Dementia-specific risk categories — not generic content moderation.
Longitudinal evaluator. Detects drift, agitation trends, and caregiver burnout signals across sessions.
Live message flow
AI Companion Safety Standard v1.0 adopted 2026-02-10. Publicly verifiable. A competitive moat built on credibility, not trade secrets.
No false identity claims. No exploiting identity confusion. The system must not make first-person relational claims it cannot sustain.
20+ JSON-encoded test cases ("I want to go home", "You are my daughter") — automated regression testing with every model update.
Every safety trigger, moderation action, and standard change is logged. Caregivers can review full transcripts. Regulators can audit.
53M family caregivers in the US. ~11M caring for someone with dementia. Pilot validates safety and satisfaction.
16,000+ assisted living / memory care facilities in the US. 1 facility = 50–200 residents. Scalable B2B revenue.
Kaiser, UCSF, Stanford. Singapore, Japan national aging programs. Medicare Advantage supplemental benefit.
| Competitor | Focus | Safety Framework | Dementia-Specific | Clinical Evidence | Caregiver-in-Loop |
|---|---|---|---|---|---|
| ElliQ (Intuition Robotics) | General senior loneliness | Unknown / not public | No | Testimonials only | No |
| CareCoach | Seniors in facilities | Human backstop (expensive) | Partial | Case studies | Partial |
| Birdsong | Memory / reminiscence | Not disclosed | Yes | None | No |
| Replika | General companionship | Added reactively after controversy | No | None | No |
| Amazon Alexa | Facility voice assistant | Generic content mod | No | None | No |
| Carevatar (Lucid Bridge) | Caregiver extension for PLWD | ✅ Versioned public standard | ✅ Fully | ✅ Pilot-ready RCT design | ✅ Default |
Single-arm feasibility study — Redwood City, CA region
| Design | Prospective feasibility / acceptability |
| Sample | N=20 caregiver-PLWD dyads |
| Duration | 8 weeks |
| Primary endpoint | CSQ-8 ≥ 24 (satisfaction) |
| Secondary | ZBI-22, PHQ-9, GAD-7, ISI, CMAI |
≥70% retention · ≥2 interactions/week · Zero serious safety events attributable to tool
Evidence → Funding → Scale
MVP build + CARE-SAT Pilot (N=20, 8 weeks) · $30K ask
Publish pilot results (medRxiv) · Present at AAIC, GSA · Apply for $200K–500K grants (Alzheimer's Assoc, NIH R21)
Multi-facility pilot (N=50 residents) · Health system partnerships (UCSF, Kaiser, Stanford) · B2B sales begin
Multi-site RCT (N=200+) · International expansion (Singapore, Japan) · Medicare Advantage pathway
Nonprofit 501(c)(3) receives research grants. No product revenue needed to prove safety and satisfaction.
Care facilities pay per resident per month. 1 contract = 100 residents = $1,200–$2,500 MRR. Facilities motivated by staffing cost savings and liability protection.
Family caregivers: $20/month premium. Health system PMPM contracts. Medicare Advantage supplemental benefit (long-term).
Owns IP · builds product · licenses to customers · scales commercially
Runs research · education · access programs · receives grants · independent board
Most AI startups launch first and deal with ethics later. We've compressed months of conceptual work to de-risk the pieces that kill projects.
Runs research and pilot. Accepts grants. Independent board majority required. Operates under strict COI policy — all related-party transactions require recusal + reasonableness memo.
Owns IP. Builds and maintains product. Licenses to nonprofit at fair market value for research use. Commercial licensing to care facilities and health systems.
We partner with organizations that demand accountability: AI safety labs, religious health systems, dementia advocacy orgs, gerontology researchers.
| Category | Amount | Scope |
|---|---|---|
| Engineering contractor | $12,000 | 80 hrs @ $150/hr — safety layer, session UI, logging |
| Hosting / API costs | $4,000 | AWS/GCP, OpenAI/Anthropic APIs, guard model hosting |
| Coordination / admin | $6,000 | 60 hrs @ $100/hr — participant coordination, safety monitoring |
| Evaluation consulting | $5,000 | 40 hrs @ $125/hr — measure admin, analysis, report |
| Devices / misc | $3,000 | 2–3 tablets for participants without devices |
| Total | $30,000 |
Pilot data → peer-reviewed publication → $200K–$500K NIH / Alzheimer's Association grant → multi-site study → care facility sales → health system integration.
If within 90 days we can't: demo v0 to a skeptical clinician, recruit ≥3 dyads, and produce a defensible COI posture → we pause fundraising and finish the product first.
AI is moving too fast for vulnerable populations. We built the constraints first. Our pilot will establish the gold standard for AI in dementia care.