IWMS - Integrated Waste Management System

The problem

Municipal waste collection followed fixed schedules regardless of actual bin fill levels, leading to wasted trips and overflow incidents.

Architecture

Sensor data ingested into a normalised SQL store, ML.NET regression predicting fill rates per bin, GIS layer overlaying optimal routes on a map UI.

Key decisions

GIS rendering on the client side rather than server-rendered tiles — faster interaction and lower server cost. Predictive model retrained weekly via a scheduled job.

Outcomes

• Reduced unnecessary collection trips through prediction-driven scheduling.
• Operational dashboards adopted by route planners as their primary tool.

What I would do differently now

Next iteration would explore reinforcement-learning approaches for route optimisation rather than static prediction plus heuristic routing.