A smart civic reporting platform that helps citizens report waste incidents and helps municipal officers review, prioritize, group, map, and manage reports.
Python · Streamlit · Gemini · Supabase · PostgreSQL · Folium
AI supports decision-making. Final municipal decisions remain with human officers.
CleanSight AI improves the way civic waste incidents are reported and reviewed.
Citizens can upload an image, provide a description and location, and receive AI-assisted analysis before submitting the report.
Municipal officers can then review reports, correct AI-generated information, identify priority incidents, manage related reports, view incidents on a map, and export report data.
- Report waste as a guest or signed-in user
- Upload waste images
- Add a simple everyday address
- Use GPS/current location
- Add landmarks and descriptions
- Run AI-assisted waste analysis
- Review the generated analysis
- Submit reports to the municipality
- Secure municipal sign-in
- Dashboard with submitted reports
- Search and filtering
- Interactive report map
- Review AI-generated analysis
- Correct report information
- Add municipal notes
- Update report status
- Detect and group related reports
- Export report data as CSV
CleanSight AI uses Google Gemini multimodal AI to analyse uploaded waste images.
The analysis can include:
- waste type
- estimated waste volume
- visible hazards
- hazard score
- road accessibility
- recommended collection vehicle
- AI confidence
The AI output is treated as decision support, not as a final municipal decision.
The current priority score considers:
| Factor | Weight |
|---|---|
| Hazard | 70% |
| Waste Volume | 30% |
Priority levels:
Score >= 75 → HIGH
Score >= 45 → MEDIUM
Otherwise → LOW
Potentially related or duplicate reports are handled on the municipal side.
Municipal officers can:
- compare nearby reports
- group reports belonging to the same incident
- mark reports as separate
- remove reports from an existing group
Original citizen reports are always preserved.
Click a section below to view its screenshots.
🏠 Landing Page
Citizens can use the guest reporting option or create an account through the application.
Use the following account to access the municipal demonstration dashboard.
Municipal Email: [email protected]
Password: municipal123
Citizen / Municipal Browser
│
▼
Streamlit App
/ \
▼ ▼
Supabase Gemini
Auth / DB Multimodal AI
Storage
│
▼
PostgreSQL
Citizen uploads report
↓
AI analyses waste image
↓
CleanSight AI calculates priority
↓
Citizen reviews and submits
↓
Report stored in Supabase
↓
Municipal officer reviews
↓
Officer confirms or corrects report
Current report lifecycle:
SUBMITTED → REVIEWED
Future versions may extend this to:
SUBMITTED → REVIEWED → ASSIGNED → IN PROGRESS → COMPLETED
| Area | Technology |
|---|---|
| Web Application | Streamlit |
| Programming Language | Python |
| AI | Google Gemini Multimodal |
| Backend | Supabase |
| Database | PostgreSQL |
| Authentication | Supabase Auth |
| Storage | Supabase Storage |
| Mapping | Folium |
| Geocoding | Geopy |
| Data Processing | Pandas |
| Image Processing | Pillow |
| Deployment | Streamlit Community Cloud |
| Source Control | GitHub |
CleanSight-AI/
│
├── app.py
├── citizen_view.py
├── municipal_view.py
├── requirements.txt
├── README.md
├── .gitignore
│
├── auth/
│ ├── sign_in.py
│ └── sign_up.py
│
├── services/
│ ├── supabase_service.py
│ ├── auth_service.py
│ ├── ai_service.py
│ ├── location_service.py
│ └── report_service.py
│
├── assets/
│ └── cleansight_logo.png
│
└── docs/
└── screenshots/
git clone YOUR_GITHUB_REPOSITORY_URL
cd CleanSight-AIpython -m venv venv
.\venv\Scripts\Activate.ps1python -m pip install -r requirements.txtCreate:
.streamlit/secrets.toml
Add:
GEMINI_API_KEY = "your-gemini-api-key"
SUPABASE_URL = "your-supabase-url"
SUPABASE_KEY = "your-supabase-key"python -m streamlit run app.pyNever commit API keys, passwords, .env files, or Streamlit secrets to GitHub.
Recommended .gitignore:
venv/
.venv/
__pycache__/
*.pyc
secrets.toml
.env
.env.*
.vscode/
.idea/
.DS_Store
Thumbs.dbMunicipal accounts should be assigned the appropriate role through a controlled administrative process.
- AI analysis may occasionally be incorrect or uncertain.
- AI confidence is not a calibrated probability.
- Sri Lankan address matching may sometimes be inconsistent.
- Related-report detection currently relies mainly on location and time.
- The current lifecycle primarily uses
SUBMITTEDandREVIEWED. - The system is an MVP and requires further production security and evaluation.
Planned improvements include:
- stronger role-based security
- improved Sri Lankan location handling
- image/text similarity for duplicate detection
- expanded incident lifecycle
- audit history
- municipal assignment
- route optimisation
- waste hotspot analytics
- collection tracking and ETA prediction
CleanSight AI is an AI-assisted civic reporting and municipal decision-support system.
AI-generated information may contain errors and should be reviewed by municipal officers before being treated as verified information.
This project is licensed under the MIT License.
See LICENSE for details.