Problem Statement
Every habagat, NCR floods gutter-to-knee while commuters guess: Waze doesn’t know lubog, MMDA posts are minutes late, and “passable to light vehicles” means different things per street. During SONA rains and Inday’s monsoon surge, workers waded blind into chest-deep esteros. Barangays clear drains late because they don’t see which inlet failed first.
Proposed Solution
Search street or drop pin: BahaPassable shows depth (ankle/knee/waist), passability per vehicle (tao/motor/kotse), and 1-hour trend from rain + upstream reports. Citizens upload 1 photo; vision estimates depth from tire/steps reference. Tanods confirm with one tap. SMS query works with no load: “BAHA Malabon Simeon” returns the latest reading.
Target Market
- Primary users: Commuters, riders, jeep drivers, parents fetching kids in NCR, Bulacan, Cagayan.
- B2G: MMDA, barangay flood marshals, school suspension deciders.
How AI Is Used
- Depth-from-photo: Vision estimates waterline against wheels, steps, bollards.
- Trend forecaster: Tiny time-series on rain + upstream depths predicts next-hour rise.
- Dedup: Embeddings merge 30 photos of the same corner into one reading.
PLDT / Smart / ePLDT / Rokid Angle
- Smart: Zero-rated SMS queries + cell-broadcast links during habagat.
- ePLDT: Hosts flood-reading archive for LGU planning.
- Rokid: Marshals see depth overlay on patrol (stretch).
Demo Scenario (3 minutes for judges)
- Search mock “Malabon MH Del Pilar”; map shows knee-deep, motor-only.
- Upload dashcam photo; AI depth appears in 5 seconds.
- SMS demo on a judge’s phone returns the same reading.
- Trend arrow shows rising next hour — “Umiwas na.”
- Close: “Bago lumusong, mag-check muna.”
30-Hour Build Plan
| Hours | Task | Role |
|---|---|---|
| 0–3 | Gather MMDA sample reports + mock photos | All |
| 3–12 | Depth estimator + trend model | AI lead |
| 3–10 | Map + SMS query | Backend |
| 10–20 | Photo upload + confirm UI | Frontend |
| 20–26 | Street-test with synthetic floods | All |
| 26–30 | Pitch, disclosure | PM / Pitch |
Suggested stack: Next.js + Leaflet, FastAPI, YOLO depth estimator, SMS gateway, Postgres. Public data / APIs: MMDA flood posts, PAGASA rain, Project NOAH.
Risks & Privacy Notes
- Safety: Never encourages crossing; “Do not cross waist-deep” hard-coded. Demo streets fictional; no real plates/faces into AI tools.
- Privacy: Faces/plates auto-blurred; location rounded to street segment.
- IP: Team mock photos + open data only.
AI Use Disclosure (draft)
Uses vision depth estimation and a small forecaster plus LLM SMS wording. Coding assistants helped with map code. All demo depths human-verified.
Rubric Scoring
| Criterion | Weight | Score (1–5) | Why |
|---|---|---|---|
| Innovation & creativity | 25% | 4 | Street passability in plain Tagalog is overdue |
| Problem–solution fit | 25% | 5 | Daily habagat pain |
| AI application & value | 20% | 4 | Depth-from-photo essential |
| Feasibility | 15% | 4 | Very demoable |
| Impact & scalability | 15% | 4 | NCR-wide with SMS |
| Weighted total | 4.30 |