LindolReady: Snap your house, know your earthquake risk before the 'Big One'

Problem Statement

PHIVOLCS warns that the West Valley Fault, which runs through Metro Manila and nearby provinces, can produce a magnitude 7.2 earthquake. The Metro Manila Earthquake Impact Reduction Study (2004) estimated such a quake could kill tens of thousands of people and collapse or heavily damage a large share of residential buildings. Recent quakes in Abra (2022), Cebu, and Davao Oriental (2025) showed again that informal, non-engineered houses are the most likely to collapse. Many Filipino homes are self-built with thin hollow blocks, weak column-beam joints, and ground-floor garages or sari-sari stores that create soft stories. Homeowners don’t know whether their house is at risk, and LGUs have no building-level vulnerability data for planning.

Proposed Solution

LindolReady is a mobile web app and Messenger bot. A homeowner follows a guided capture of 4–6 photos (front facade, corners, ground floor, columns and walls up close, roof) and answers a short checklist (number of storeys, year built, whether an engineer was involved). A vision model and multimodal LLM detect risk features: soft or weak stories, short columns, irregular shapes, hollow-block thickness and visible cracks, overhangs, and heavy roofs or water tanks. These are combined with location-based hazard layers from HazardHunterPH (distance to active faults, liquefaction, and ground-shaking intensity). The result is a simple 0–100 vulnerability score with color bands and explanations in plain Filipino. It also gives a prioritized list of retrofit tips with rough peso cost ranges and links to accredited engineers and LGU or Pag-IBIG housing programs. Anonymized scores feed an LGU dashboard heatmap to target inspections, retrofitting subsidies, and evacuation planning.

Target Market

• Primary users: Homeowners and renters in earthquake-prone urban areas, starting with the West Valley Fault corridor (Metro Manila, Bulacan, Rizal, Cavite, Laguna) and extending to Cebu, Davao, and Baguio.
• Buyers (B2G): City DRRMOs, building officials (National Building Code enforcement), DHSUD, and MMDA.
• B2B: Insurers (earthquake cover pricing), banks and Pag-IBIG (housing loan risk), hardware chains and contractors (retrofit leads), and schools assessing campuses.
• Size: Millions of households in Metro Manila and nearby fault-adjacent provinces.

How AI Is Used

• Vision feature detection: Identifies structural red flags from ordinary phone photos. Without it, assessment needs an engineer site visit that most households can’t afford or schedule.
• Multimodal LLM reasoning: Combines photo findings, checklist answers, and hazard layers into an explainable score using a rule-based framework adapted from FEMA P-154 rapid visual screening, so outputs aren’t a black box.
• LLM retrofit advisor: Produces tailored, plain-language, cost-tiered recommendations in Filipino and English.
• Spatial aggregation: Clusters scores into barangay- and street-level heatmaps for LGU prioritization.

PLDT / Smart / ePLDT / Rokid Angle

• Smart subscriber reach: SMS and Messenger campaigns during Earthquake Drill season (the quarterly Nationwide Simultaneous Earthquake Drill) to drive mass sign-ups.
• ePLDT cloud and data centers: In-country hosting of the LGU risk database and AI inference, sold as a B2G SaaS.
• PLDT Enterprise B2G/B2B: Bundle for LGUs (heatmap dashboard) and insurers/banks (risk scoring API).
• Rokid AR (stretch): LGU building inspectors wear Rokid glasses for hands-free rapid visual screening, using the same model to pre-fill FEMA P-154-style forms.

Demo Scenario (3 minutes for judges)

1. Open with a PHIVOLCS West Valley Fault map. “This line runs about 2 km from DLSU. Is your house ready?”
2. A teammate uploads five photos of a real (team member’s, with consent) two-storey house with a ground-floor sari-sari store.
3. LindolReady highlights the open ground floor (“soft story: high risk”) and thin hollow-block walls, adds the fault distance and liquefaction layer, and outputs Score 72/100: High vulnerability.
4. It shows three retrofit tips with peso ranges (e.g., add infill walls or steel bracing at the store opening, anchor the water tank, consult a licensed engineer) in Filipino.
5. Switch to the LGU view: a heatmap of a sample barangay built from synthetic data shows red clusters along the fault. “Now the city knows where to send its engineers first.”

30-Hour Build Plan

Suggested stack: Next.js PWA, a multimodal LLM (Gemini/GPT-class) plus an optional YOLO classifier, FastAPI, PostGIS, Leaflet/Deck.gl heatmap, Messenger Platform API. Public data / APIs: PHIVOLCS Active Faults and FaultFinder, HazardHunterPH/GeoRiskPH (ground shaking, liquefaction), Metro Manila Earthquake Impact Reduction Study summaries, FEMA P-154 RVS methodology, PSA census housing characteristics (construction materials), OpenStreetMap building footprints.

Risks & Privacy Notes

• Not an engineering certification: Show clear disclaimers, use conservative scoring, and always recommend a licensed civil engineer for high scores. Wrong “safe” ratings are the biggest risk, so bias the model toward flagging problems.
• Photo quality and hidden structure: Use guided capture prompts, show confidence indicators, and ask checklist questions about things photos can’t show (rebar, foundation).
• Data Privacy Act 2012: House photos and addresses are personal data. Get explicit consent, let users delete their data, send only aggregated, coarse-grained cells (not exact addresses) to LGU heatmaps unless the user opts in, and encrypt storage.
• Misuse by insurers/lenders: Data is shared with third parties only with explicit user consent, never sold.
• Clause 24 compliance: Use only team-owned photos (with consent), public sample images, and synthetic heatmap data in the demo. No third-party personal data goes into AI tools.

AI Use Disclosure (draft)

LindolReady uses a multimodal LLM (e.g., Gemini/GPT-class) and optional vision classifiers to detect structural vulnerability features, an LLM to generate retrofit advice, and spatial clustering for heatmaps. AI coding assistants helped with scaffolding and drafting this document. Scoring rules are based on the published FEMA P-154 methodology and reviewed by the team. All outputs carry disclaimers and recommend professional engineering review.

Rubric Scoring