ElderCare

Inspiration

Taiwan became a super-aged society in December 2025: 20.7% of people are now over 65. Taipei is the oldest city (24.18% aged 65+), yet it ranks 20th nationally in long-term care coverage (58.1%). The help exists. Long-Term Care 2.0/3.0 is generous. But families can't navigate it.

When a parent falls, has a stroke or comes home from hospital, the adult child suddenly faces 4 benefit types, 7 care levels, 3 copay groups, prior-use rules and 11,722 government-contracted providers. Most families lose weeks before the first call. Some never use benefits they're entitled to. Others let money expire without knowing it.

What it does

ElderCare turns that system into an exact plan in about 2 minutes:

Eligibility, stated honestly. It tells families whether their parent meets the conditions to apply (65+, Indigenous 55+, dementia, disability certificate, post-acute care). It never guesses a care level: the government home assessment decides that.
Exact money. It shows what the government pays and what the family pays, using the official 2026 caps and copay rates: $$\text{you pay} = \text{cap}{\text{level}} \times r{\text{household}}$$ For example, at level 4 a general household gets NT\$18,580/month of care and pays NT\$2,973.
What's already used, and what's about to expire. Unused care money resets 6 months after the care-plan approval month. Device budgets run on 3-year periods. Respite is a yearly amount. ElderCare calculates what's left and warns before money resets.
Who to call first, and what to say.
- Still in hospital: ask the ward nurse for discharge planning. It gets families into care in about 4 days instead of 51.
- Not yet applied: the city's long-term care center and its online form.
- Already assessed: the nearest case-management center.
- Each case comes with a ready-to-read call script.
Real providers near you. Contracted providers that serve the district, care homes with open beds, and community care centers that need no assessment at all.

Every number links to its official source.

How we built it

Built in Lovable. We built the whole app in Lovable, one feature per prompt:
- Project Knowledge held our product brief, design system and rules.
- Plan mode handled the risky steps (data layer and rules engine).
- Lovable Cloud hosts it.
- Lovable AI writes a warm plain-language summary on top of the computed plan. The plan still shows if the AI fails.
- The app is React, TypeScript, Tailwind and Leaflet maps, published straight from Lovable.
Real government data. We downloaded the Ministry of Health and Welfare's national provider file, 長照ABC據點 (data.gov.tw #88270, updated daily), and cleaned it with a small Python script:
- merged 26,240 rows into 11,722 providers;
- decoded government area codes into districts;
- grouped about 15 service names into 10 categories in Chinese and English;
- calculated open beds.

We also added New Taipei's 648 community care centers from the city's live open-data API. The cleaned files are loaded into the Lovable app as static data.

Rules as data:
- We read the 2025-amended 長期照顧服務申請及給付辦法 and the 1966.gov.tw benefit pages.
- We encoded them as a structured rules file: caps, copays, carry-over, device-period rules, respite prices, residential subsidy and tax deduction.
- Every rule links to its source.
Deterministic rules engine: pure TypeScript, built in Lovable. All NT\$ amounts come from code, never from AI.
Design: mobile-first, Traditional Chinese with an English toggle, no login. Answers stay in the browser session.

Challenges we ran into

Turning law into logic. Copay categories, the 6-month carry-over and the 2026 device transition rules (group 1 users wait 3 years for the new NT\$60,000 group) are written in legal Chinese across several documents. We cross-checked them against the official pages.
Saying only what's true. We removed every "you likely qualify" and designed around the two real gates: conditions to apply, then the official assessment.
Messy open data. It came with coded districts, duplicate rows per service, mixed service names and some missing coordinates.
Prompting for precision in a 3-hour build. We used Project Knowledge, Plan mode and one-feature-per-prompt so Lovable never invented numbers and never broke working parts.

Accomplishments that we're proud of

Real government data only, no mock data:
- 2,666 providers in Taipei and New Taipei;
- 11,722 nationwide, already cleaned;
- 197 care homes with live open-bed counts.
The money-expiry warning. No existing tool shows it, and it's real money families lose.
One source link per rule, so judges, families and case managers can check us.

What we learned

Taiwan's long-term care system is strong; the gap is navigation.
"Government" care is mostly private. 96% of contracted providers are private companies, clinics and nonprofits; the government pays.
AI is most useful when it explains numbers it didn't make. In Lovable, that meant keeping the math in code and the AI on top.

What's next for ElderCare

A LINE bot, since that's where Taiwanese families already are.
Partnering with hospital discharge planners, so families get the plan before they leave the ward.
Nationwide coverage, using the cleaned national file for all 22 cities and counties.
Community care, to help neighbors and volunteers join the care network. By 2035, 1.3 million people in Taiwan will need long-term care.