Hearthline — 90-Day MVP Roadmap
Success is not “AI platform.” Success is: a Portmore household photographs real clinic paper, shares a radar into WhatsApp, and nobody was asked for a card.
Team (small, Jamaica-present)
- 1 product/design (must be on-island part of each week)
- 1 full-stack
- 1 ML/extract (can be remote; tested on Jamaican paper)
- 1 community (Kingston + one rural parish)
- Legal: DPA notice, company registration, 876 number
Milestones
Days 1–21 — Trust before pixels
- Register the Jamaican entity; print the red-flag card.
- Collect 40 real (consented) anonymised discharge/script photos from two parishes.
- PWA shell: shutter, trust copy, 876 footer.
- Metric: 10 strangers will photograph a *dummy* paper without asking “ah weh unuh tek mi card number?”
Days 22–50 — Extract + radar
- Confirm/correct loop that is faster than typing.
- Polar radar; share PNG; revoke.
- Offline last-radar.
- Metric: on a 20-document holdout, ≥70% of high-blast-radius intervals recoverable with ≤2 corrections.
Days 51–75 — Parish desks
- Two walk-in partners (pharmacy or credit union) in Kingston Metro and one other parish.
- Diaspora view link tested from a US number.
- Metric: 30 live households, 0 payment-phishing reports.
Days 76–90 — Quiet launch
- Circle billing only for overseas volunteers, after free view.
- Harbor mode.
- Kill criterion: if people think we take remittance, we stop marketing and rewrite every screen.
Resource requirements
- Phones: mid-range Android (Redmi/Tecno class), not a demo iPhone-only.
- Data: Digicel + FLOW SIMs for field tests.
- Budget honesty: the expensive part is being present, not GPUs.
Success metrics (90 days)
| Metric | Target |
|---|---|
| Households that completed one confirmed radar | 100 |
| Shares opened by a second person | 60 |
| Island users asked to enter a card | 0 |
| Reports of fake-Hearthline Lynk asks | 0 (and a public 876 to report clones) |
| Median time, photo → share | < 4 minutes |
Out of scope (90 days)
Helper marketplace, hospital FHIR, iOS-only polish, tokenisation of medical records.