LindolLens: Phone-photo crack triage after the Cebu and Mindanao quakes

Problem Statement

The Sept 2025 Cebu M6.9 and June 2026 Mindanao M7.8 damaged schools, roads, and boarding houses; principals suspended classes “until further notice” while waiting days for an engineer. Families sleep outside over hairline cracks or, worse, sleep inside under a shear crack. LGU engineers cover hundreds of buildings with clipboards.

Proposed Solution

Snap 3 photos (whole wall, 1m, close-up) + answer 4 taps (year built, floors, shaking felt, aftershock damage). Vision AI grades crack type (hairline, diagonal shear, column spalling) and an LLM writes a plain-language advisory: “HUWAG GAMITIN — tumawag ng engineer ngayon” vs “OBSERBAHAN — kuhanan ulit sa 48h.” Queue ranks schools and boarding houses for the municipal engineer with map pins.

Target Market

• Primary users: Principals, dorm owners, parents in quake-hit Cebu, Davao, GenSan, Sarangani.
• B2G: DepEd division engineers, MDRRMOs, DPWH damage teams.

How AI Is Used

• Vision crack classifier: Distinguishes cosmetic vs structural patterns; essential because text descriptions fail.
• LLM triage writer: Converts labels + building age into Bisaya/Tagalog actions with confidence.
• Priority ranker: Combines severity, occupancy (classroom vs storage), and aftershock reports.

PLDT / Smart / ePLDT / Rokid Angle

• Smart: SMS fallback when data is down; post-quake cell-broadcast links to triage.
• ePLDT: In-country hosting of school damage vault for DepEd audits.
• Rokid: Engineers see crack overlay + history on-site (stretch).

Demo Scenario (3 minutes for judges)

1. Show a mock classroom wall with diagonal shear photo; app verdict: “CUT”
2. Show hairline paint crack; verdict: “OBSERVE — 81% cosmetic.”
3. Map ranks 3 mock schools for tomorrow’s inspection.
4. Bisaya voice readout plays for lola caretaker.
5. Close: “Bago matulog sa loob, picturan muna.”

30-Hour Build Plan

Suggested stack: Flutter + YOLO crack detector, LLM explainer, SQLite queue, Leaflet school map. Public data / APIs: PHIVOLCS guides, open concrete-crack datasets, news damage photos (reference only).

Risks & Privacy Notes

• Life-safety: Always “AI lead only — engineer decides”; low-confidence defaults to stay-out. Demo uses mock walls, no real addresses into AI tools.
• Privacy: Photos stay local until user consents to share with LGU; faces auto-blurred.
• IP: Open datasets + team photos only, licenses listed.

AI Use Disclosure (draft)

Uses vision classification plus LLM advisories; AI coding assistants helped with mobile scaffolding. All demo verdicts checked against PHIVOLCS guidance by the team.

Rubric Scoring