Pothole

Inspiration

Over 25% of U.S. roads are rated “poor” or “mediocre,” costing drivers over $26 billion in repairs every year. Yet, most cities still depend on slow, manual road inspections. We wanted to create a scalable, data-driven system that could automatically detect and map road hazards—empowering both the public and government to make streets safer.

What it does

Pothole is a two-sided platform that uses AI and Google Street View imagery to map road conditions in real time.

• For the public: Anyone can anonymously report a hazard by uploading a photo. Our AI model instantly classifies the hazard type and severity.
  
• For officials: Government users can log in to a secure dashboard to view all reports, run virtual surveys of unreported areas, and prioritize repairs using our AI-generated heatmap.

How we built it

• Frontend: React + Leaflet for the interactive map and reporting interface.
  
• Backend: Supabase for authentication, data storage, and file uploads.
  
• Machine Learning: YOLOv12 model (hosted with Dedalus Labs) trained on open-source pothole datasets to classify and score road hazards.
  
• Data Pipeline: Google Street View API for pulling road imagery and batch processing with the ML model.
  
• Visualization: Real-time heatmaps and hazard clusters for municipal dashboards.

Challenges we ran into

• API limitations from Google Street View required caching and smart batching.
  
• Dataset variability—roads differ drastically across regions, making it hard to generalize the model.
  
• Balancing accuracy and performance for real-time detection under hackathon time constraints.
  
• Integrating authentication and roles (anonymous public vs. verified officials) cleanly in Supabase.

Accomplishments that we're proud of

• Built a fully functional end-to-end prototype connecting AI inference, database, and real-time map visualization in under 36 hours.
  
• Created a secure two-role system for both public and government users.
  
• Deployed our first version of the AI model through Dedalus Labs for scalable inference.
  
• Designed a sleek, intuitive UI for both community reports and official dashboards.

What we learned

• The power of combining citizen-sourced data with machine vision to make infrastructure monitoring scalable.
  
• How to manage large-scale image processing efficiently using public APIs.
  
• The importance of thoughtful UX when serving both technical users (city engineers) and non-technical users (drivers).
  
• How cloud tools like Supabase and Dedalus Labs can accelerate full-stack AI integration.

What's next for Pothole

• Partner with municipalities to pilot our system on real maintenance schedules.
  
• Integrate with navigation apps to help drivers avoid high-risk areas.
  
• Continue training models via Dedalus Labs for continuous improvement and global scalability.

Built With

• dedalus
• google-streetview
• leaflet.js
• python
• react
• tailwindcss
• vite