TownPass Recycling
Award-winning recycling microservice built in a 24-hour Taipei City hackathon (top team of 111 nationwide) and shipped live on the official TownPass (台北通) app — LLM trash recognition + garbage-truck alerts, Express.js backend, serverless CI/CD on GCP.
Award-winning team project — the CityPass Microservice Hackathon at Taipei’s Autumn Coding Festival (2024). Our service went live on Taipei’s official TownPass (台北通) app.
TownPass Recycling is a recycling microservice built for the CityPass Microservice Hackathon, a competition run by the Taipei City Department of Information Technology to promote open-source and open-data. The event was open to all ages and judged purely on a live demo; out of 111 teams nationwide, ours stood out over two sleepless nights — and the service shipped live on Taipei’s official “TownPass (台北通)” app.
What it does
- Trash image recognition. Photograph an item and the app recognizes it, then tells you its recycling category and how to dispose of it — making sorting effortless.
- Garbage-truck tracking. Using live garbage-truck data, it provides navigation to the truck and an arrival alarm, so you never miss it again.
My role
I owned the generative-AI / image-recognition model, the entire backend, part of the frontend, and the final presentation. I also split the system into frontend / backend / model training / dev-&-deployment pipeline so the team could build in parallel.
Engineering highlights
The core challenge was shipping a working microservice in 24 hours and demoing it live to the judges. Because it was headed for the official TownPass app, we designed for the government’s priorities — fast to build, easy to maintain, low cost:
- Serverless CI/CD on GCP. We ran everything on Google Cloud Platform with Cloud Run and
Cloud Build for an automated CI/CD pipeline — no physical servers to manage. Code lived on
GitHub, with the
mainbranch protected behind Pull Requests so nothing broken could reach production during rapid development. It was the team’s first time using this workflow, and we made it work. - Backend. I built the backend on Express.js for development speed, exposing a RESTful API to the frontend.
- Image recognition. I evaluated Google Teachable Machine, Lobe, and LLMs, and chose an LLM for both recognition and generating the recycling guidance.
- Frontend. We solved a stack of device-level challenges — camera capture & upload, Mapbox maps, turn-by-turn navigation, GPS positioning, and alarm reminders.
Beyond the tech, it was a great exercise in teamwork — and a real confidence boost to place among the top teams nationally, ahead of many made up of companies and industry professionals.