Backend Case Study
Gotcha Trash
A real-time AI-powered application developed under hackathon constraints to promote environmental sustainability.
AIARHackathon
Problem
Public waste sorting relies heavily on user compliance, with little feedback or incentive for correct disposal.
Goal
Design a real-time system that detects trash and bins, provides immediate feedback, and incentivises correct behaviour.
System Architecture
Low-latency, real-time inference pipeline integrated into an AR experience.
- Camera feed captured by AR mobile application
- Pre-trained YOLOv8 model for object detection
- Inference results drive game logic and reward system
- Minimal backend to prioritise real-time user experience
Key Decisions & Trade-offs
- Used pre-trained models to maximise speed of development during the hackathon
- Prioritised on-device inference to reduce network latency
- Kept backend minimal to focus on interaction and performance
Tech Stack
Unity · Python · YOLOv8 · OpenCV · TensorFlow · ARKit
Links
Live demo (WIP)