A.KTP
← Back to projects
AndroidAI/MLBackend

Grama Waste Tracker (PashuSwasthya)

A village waste-management Android application paired with a full AI-integration architecture for classification, forecasting, and route optimization.

View on GitHub Discuss this project

Overview

Grama Waste Tracker is an Android application built to give villages a digital system for tracking and managing waste collection, developed during an Android development internship. Alongside the app, I designed a full AI-integration architecture to make the system genuinely intelligent rather than just a digital logbook.

Problem statement

Villages lacked any structured, digital way to track waste generation, classify what was being collected, or plan collection routes efficiently. Collection was largely manual and unoptimized, with no forecasting or systematic classification in place.

Architecture

The Android app handles data capture and tracking in the field. Behind it, the designed architecture adds three AI capabilities: computer-vision-based waste classification, demand forecasting, and collection-route optimization — backed by a FastAPI service and PostgreSQL database, containerized for deployment.

  • Android app (Kotlin) for on-the-ground data capture and tracking
  • YOLOv8 model for image-based waste classification
  • Prophet time-series model for waste-volume demand forecasting
  • OR-Tools for optimizing collection routes
  • FastAPI + PostgreSQL backend, containerized with Docker
Android app FastAPI backend PostgreSQL YOLOv8 classification Prophet forecasting OR-Tools route optimization

Tech stack

KotlinAndroid StudioFastAPIPostgreSQLYOLOv8ProphetOR-ToolsDocker

Challenges & solutions

  • Field usability — the app needed to work for non-technical users in the field, so the Android UI was kept deliberately simple and task-focused.
  • System scope — rather than bolting AI on as an afterthought, the classification, forecasting, and routing components were designed as first-class parts of the architecture from the start.
  • Deployment readiness — containerizing the backend with Docker was a deliberate choice to make the architecture portable for scaled rollout beyond a single village.

Results

Shipped
Working Android app
3-model
AI architecture designed
Container-ready
Docker deployment

Future improvements

  • Deploy the YOLOv8 classification model as a live inference endpoint connected to the app camera
  • Add a dashboard for municipal staff to view forecasts and optimized routes directly
  • Pilot the system in a real village deployment to validate the forecasting accuracy against actual collection data