A commercial building prospecting engine for Grundfos water reuse systems. The app identifies high-value buildings across US cities by combining satellite imagery analysis, financial data, and regulatory signals into a single Viability Score surfaced through an interactive map.

Built for the Grundfos sponsor track at HackSMU VII. Devpost

• Lets you run a live audit on any city and filter/explore/save results in an interactive map
• Scores each building on roof area, local water cost, precipitation, stormwater fees, ESG commitments, tax incentives, and flood risk
• Detects cooling towers in satellite imagery using a two-stage computer vision pipeline rendered as precise map markers

Frontend

React (Vite), MapLibre GL, Deck.gl, Zustand, Mantine

ML / CV microservice

YOLOv5, EfficientNet-B5, PyTorch, Flask

Data

• Microsoft Overture Maps — building footprints and attributes
• Google Maps Static API — satellite tiles for CV inference
• World Population Review — water cost by state/city
• NOAA / Open-Meteo — precipitation data
• Science Based Targets initiative (SBTi) — ESG commitment lookup
• FEMA / local municipal sources — flood risk, stormwater fees

npm install
npm run dev

pip install -r requirements_towerscout.txt
python tower_server.py --yolo-weights yolov5_best.pt --en-weights b5_unweighted_best.pt

Based on “Automated cooling tower detection through deep learning for Legionnaires’ disease outbreak investigations” (Wong et al., 2024, The Lancet Digital Health).

• Methodology adapted from the TowerScout project (UC Berkeley / CDC).
• YOLOv5 model retrained using reconstructed pipeline and partial dataset.
• EfficientNet pretrained weights used.

Original work licensed under CC BY-NC-SA 4.0.