SolarSense: AI that shows every LGU which public roofs can keep the lights on

Problem Statement

The Philippines has some of the highest electricity rates in Southeast Asia, and grid yellow and red alerts during the hot dry months, along with typhoon-caused outages, regularly cut power to schools, health centers, and evacuation sites. Many public schools and barangay halls double as evacuation centers, yet they go dark exactly when communities need them most. The country enjoys strong solar irradiance, and the Renewable Energy Act (RA 9513), net metering, and DOE programs encourage adoption. Most LGUs, however, lack the engineers and data to judge which buildings are worth solarizing, how big each system should be, and what the payback is, so funding stalls at the feasibility stage.

Proposed Solution

SolarSense LGU is a web planning tool. The LGU picks its territory. The system pulls the locations of public facilities (DepEd schools, barangay halls, RHUs and hospitals) from open data and OpenStreetMap, then uses a vision model on satellite imagery to segment each roof, estimate usable area, orientation, and shading, and subtract obstructions. It combines this with solar irradiance (Global Solar Atlas/NASA POWER), local electricity tariffs, and typical load profiles to estimate kWp capacity, annual kWh, CAPEX, payback years, and CO₂ avoided. It also sizes a battery for critical loads (vaccine refrigerators, lights, communications) during brownouts. A resilience score weights evacuation centers and facilities in hazard zones. An LLM writes a project brief ready for LGU budget, DOE, or donor proposals.

Target Market

• Primary users: City and provincial planning and development offices, LGU engineering offices, and local DRRM offices.
• B2G buyers: LGUs, DepEd (school electrification), DOH (health facility resilience), DOE, and DILG.
• B2B/partners: Solar EPC installers, energy-as-a-service providers, banks offering green loans (e.g., LANDBANK, DBP), and development partners.
• Size: 1,600+ LGUs. Tens of thousands of public schools and around 42,000 barangay halls, plus RHUs and hospitals.

How AI Is Used

• Computer vision (roof segmentation) finds roof outlines, usable area, and obstructions from satellite images. Without it, every building needs a manual site survey.
• Orientation and shading estimation from roof geometry and surrounding objects refines yield. Simple area-times-irradiance estimates overstate output.
• Forecasting/simulation: ML-assisted energy yield and load modeling uses irradiance and typical consumption to size PV and batteries.
• Multi-criteria ranking balances ROI, resilience (evacuation-center role, hazard exposure), and equity (off-grid or GIDA facilities).
• LLM generates proposal-ready project briefs and answers planner questions like “Which 10 schools give the fastest payback under ₱5M?”

PLDT / Smart / ePLDT / Rokid Angle

• Resilient connectivity: solarized schools and halls can host Smart cell sites, free WiFi hotspots, and emergency communications that stay online during outages. This lowers PLDT/Smart’s own backup power costs.
• ePLDT cloud hosts imagery processing and the LGU SaaS dashboard. Later, Smart IoT can add real-time solar performance monitoring.
• PLDT Enterprise B2G can bundle solar-plus-connectivity resilience packages for LGUs, DepEd, and DOH.
• ESG alignment with PLDT’s own sustainability and renewable energy targets.
• Rokid AR glasses let installers and LGU engineers see proposed panel layouts and shading overlays on the actual roof during site validation.

Demo Scenario (3 minutes for judges)

1. Opening image: a barangay health worker in Tacloban holding vaccines in a warming cooler during a post-typhoon blackout.
2. Select “Tacloban City” on the map. More than 100 public facilities load as pins.
3. Click a public elementary school. The AI outlines its roofs from satellite imagery and reports “620 m² usable → 85 kWp, ~115 MWh/yr, payback ~5–6 years, 60 tCO₂ avoided/yr.”
4. Toggle “Resilience mode.” Evacuation centers and RHUs in flood or storm surge zones jump to the top, with battery sizing for 24-hour critical loads.
5. Ask the copilot: “Top 10 facilities for a ₱20M budget.” A ranked list and a downloadable LGU project brief appear.
6. Close: “Every barangay hall and school can become a power and connectivity lifeline.”

30-Hour Build Plan

Suggested stack: Python, Segment Anything / U-Net, rasterio, pvlib, GeoPandas, FastAPI, React + Leaflet/Mapbox, Gemini or an open LLM, WeasyPrint for PDFs. Public data / APIs: Sentinel-2 and openly licensed high-res tiles (e.g., Esri/Mapbox within terms), Google Open Buildings, Global Solar Atlas, NASA POWER API, PVGIS, DOE renewable energy and tariff data, Meralco/electric co-op published rates, DepEd school location data (data.gov.ph), OpenStreetMap, HazardHunterPH and Project NOAH hazard maps.

Risks & Privacy Notes

• Imagery resolution: free imagery may be too coarse for small roofs. Use footprint layers plus SAM on the best licensed tiles and show uncertainty ranges.
• Estimates are not engineering designs: results are pre-feasibility only. Structural checks, roof condition, and site surveys are still required. Label this clearly.
• Tariff and cost assumptions vary by utility and co-op. Make them editable.
• Data Privacy Act of 2012: only public facilities and non-personal data are used. No private homes are analyzed or displayed.
• Clause 24 compliance: no confidential LGU data goes into AI tools. Imagery and data licenses (Open Buildings CC BY 4.0, OSM ODbL, provider terms of service) are respected, and AI use is disclosed.

AI Use Disclosure (draft)

We use Segment Anything / open building segmentation models for roof detection, pvlib (non-AI physics model) for yield, and Gemini/an open LLM for the planner copilot and project brief generation. ChatGPT and GitHub Copilot helped with code scaffolding and pitch drafting. Team members reviewed and validated all estimates, generated briefs, and code against reference installations and published data.

Rubric Scoring