AI가 매일 어르신께 전화하고, 가족에게 건강 리포트를 보내드립니다.
Competition entry — 공군창업경진대회 (Air Force Startup Competition), 2026 My role: 팀장 · AI & Architecture — I led the team and designed the AI call / voice-analysis architecture.
CareRing is an AI-powered wellness call platform that places regular phone calls to elderly parents on behalf of their adult children, analyzes voice and speech patterns for health signals, and delivers weekly health reports back to the family.
Korea entered super-aged society in 2025 — over 20% of the population is now 65+. More than 2.2 million elderly people live alone, and roughly 3,600 lonely deaths (고독사) occur each year, with an average of 4.2 days passing before discovery. Existing solutions are either government-run programs with no family visibility, or expensive, device-dependent products most families won't adopt.
Domestic digital-care market sizing: TAM ₩1.2T → SAM ₩360B (households with an elderly parent living alone, guardians aged 40–60) → SOM ₩12B (50,000 households within 3 years).
One-liner: AI가 매일 어르신께 전화하고, 가족에게 건강 리포트를 보내드립니다. (The AI calls your elderly parent every day and sends your family a health report.)
The guardian sets a call schedule, the AI calls the parent on any existing phone (no app or device needed on their end), each call is analyzed for health/emotional signals, and the guardian receives a plain-language report.
Four-step service loop: 보호자 설정 (guardian setup) → AI 발신 통화 (AI places the call) → 음성 분석 (voice analysis) → 리포트 전송 (report delivered).
As team lead, I owned the technical core of CareRing — the part that turns "a phone call" into a health signal. This built directly on my prior work building an emotion-analysis system for video interviews at QCRI (Qatar Computing Research Institute), adapted here from video to voice-only, real-time conversation.
I designed:
- The voice health pipeline — extracting physical indicators (response latency, speech rate in WPM, pause frequency) and emotional/cognitive trend signals (embedding-based mood tracking) from each call, without relying on any wearable or app
- The AI call engine — real-time Korean conversation via LLM + TTS/STT, tuned to keep an elderly speaker comfortable and talking naturally rather than answering a survey
- The escalation path — keyword-based emergency detection mid-call, triggering an immediate push alert to the guardian
- The "wellness, not diagnosis" framing — a deliberate constraint: the system tracks trends and flags changes, and avoids medical/diagnostic language, both for regulatory reasons and to keep the product trustworthy rather than alarming
Three-layer architecture: Guardian app (client) → call/report backend → AI call engine and voice-analysis pipeline.
| Feature | Description |
|---|---|
| AI 발신 통화 | Scheduled outbound calls in natural Korean conversation |
| 음성 건강 분석 | Physical + emotional trend tracking across 30/90-day windows |
| 주간 리포트 | Plain-language weekly report delivered to the guardian |
| 응급 키워드 감지 | Instant push alert when distress keywords are detected mid-call |
| 수신 통화 (Premium) | Elderly parent can call the AI anytime |
| 가족 공유 | Share reports with up to 5 family members |
| Plan | Price | Includes |
|---|---|---|
| 베이직 | ₩9,900/mo | 주 3회 AI 통화, 주간 텍스트 요약 |
| 스탠다드 | ₩19,900/mo | 매일 통화, 그래프 리포트, 이상 징후 알림 |
| 프리미엄 | ₩34,900/mo | 매일 통화 + 어르신 발신 무제한, 월간 PDF, 5인 공유 |
A working demo of the guardian-facing app is in demo/app.html — three tabs (홈 / 분석 / 통화 기록) with live charts rendered from sample data, no external dependencies.
- TAM ₩1.2T — domestic digital-care market
- SAM ₩360B — households with an elderly parent living alone, guardian aged 40–60
- SOM ₩12B/yr — 50,000 households within 3 years
The closest existing product is 클로바 케어콜 (Naver/LINE) — but it's B2G only: municipalities choose which elderly residents receive calls, and families get no visibility at all. CareRing is B2C: families sign up directly and are the ones who receive the reports.
Positioning matrix — access model (B2G vs. B2C) × whether the family receives a report. CareRing is the only B2C option with direct family reporting.
Planned path: MVP (0–6mo, 200-household beta) → launch (6–12mo) → B2B expansion (1–2yr) → global (3yr).
| Role | Name | Background |
|---|---|---|
| 팀장 / AI & Architecture | 정의서 | CMU CS; AI researcher at QCRI; built emotion-analysis system for video interviews |
| Web Platform | 최지현 | SW engineer; national R&D government website development |
| Security Architecture | 조성윤 | 정보보호병; 정보보안기사 certified; prior security-firm experience |
| App & Server | 홍준서 | Fullstack SW engineer; app, web, and server real-service deployment experience |
This was our team's entry to the 공군창업경진대회 (Air Force Startup Competition), 2026 — written and designed during my military service. The repo contains the full application document, supporting figures, and a working front-end demo of the guardian app and marketing site.
Positioning line from the application: "내가 매일 전화드릴 수 없을 때, 케어링이 대신합니다." — "When I can't call every day, CareRing calls for me."