Real-time Monitoring • AI Route Optimization • ML Predictive Routing
WasteWise is a full-stack smart waste management platform that replaces static collection schedules with a dynamic, data-driven approach. Bin fill levels are tracked in real-time, and an AI engine calculates the most efficient collection routes — reactively for bins that are already critical, and proactively for bins predicted to overflow using machine learning.
- Monitor: Real-time bin fill levels managed through a live admin dashboard.
- Optimize: Shortest collection paths solved with the Capacitated Vehicle Routing Problem (CVRP) using Google OR-Tools.
- Predict: Random Forest ML model forecasts which bins will cross the critical threshold in the next 24 hours, enabling proactive dispatch before overflow.
- Sustain: Fewer unnecessary trips means lower fuel consumption and CO₂ emissions.
WasteWise is a three-service system, each with a dedicated technology stack:
| Service | Tech Stack | Responsibility |
|---|---|---|
| Frontend | React 19 Vite TypeScript TailwindCSS TanStack Query TanStack Router Leaflet |
Web dashboard with Admin and Driver views |
| Backend | NestJS TypeScript Drizzle ORM Neon PostgreSQL Zod JWT |
REST API for auth, bins, vehicles, routes, and algorithm orchestration |
| Algorithm | FastAPI Python OR-Tools scikit-learn pandas psycopg2 |
CVRP solver and Random Forest predictive routing engine |
Frontend (React)
↕
Backend (NestJS) → Algorithm (FastAPI)
↕ ↕
Database (Neon PostgreSQL)
Fetches all bins with fill_level ≥ 75% and feeds them into the OR-Tools CVRP solver. A dynamic depot is placed at the average coordinate of all critical bins, and PATH_CHEAPEST_ARC finds the shortest collection sequence within the vehicle's weight capacity.
A Random Forest Regressor is trained on 90 days of historical sensor data using features like hour_of_day, day_of_week, is_weekend, and bin_id_encoded. Given any time horizon, it predicts each bin's fill increase and proactively routes the truck to bins that will exceed 75% — before they become a problem.
A /simulate-next-day endpoint adds 5–25% fill to all bins in one database transaction, allowing the system to be demoed from an empty state to a critical collection scenario in seconds.
- Live Leaflet map with color-coded bin status (green / amber / red)
- Generate reactive or predictive routes and assign them to a vehicle
- Simulation and safe-reset controls for demos
- Full management of bins, operator accounts, and routes
- View the active assigned route with ordered waypoints on a map
- Mark each collection stop as complete one by one
- Track overall route progress and summary
The main monorepo — all three services live here.
web
React 19 + Vite dashboard for admins and drivers. Uses TanStack Query for server state, TanStack Router for navigation, and react-leaflet for interactive maps.
api
NestJS REST API with five modules: auth, bins, sensors, vehicles, and routes. Authenticates with JWT, uses Drizzle ORM against a Neon PostgreSQL database, and validates all algorithm responses with Zod before persisting.
ml
FastAPI Python microservice with three domains:
/optimize-cvrp— OR-Tools CVRP solver for currently critical bins/ml/predictive-cvrp— Random Forest model predicts fill levels ahead and routes proactively/simulate-next-dayand/reset-bins— simulation controls for demos
Project documentation and research.
- Graduation Project thesis and research papers on CVRP and smart waste systems
- UML, circuit schematics, and system architecture diagrams
- Slide decks and project showcases
WasteWise is developed as a Graduation Project at Haliç University, Department of Computer Engineering.
