Bingqian Liu

A community site I built and still run. People join topic groups, post and comment, chat inside each group, message each other and keep a personal schedule with reminders. The homepage is a recommended feed, and the interface is all in English.

How it grew

I started it in January 2024 while working in China, as a small forum on a familiar stack: Spring Boot with server-rendered FreeMarker and layui pages.

After moving to Melbourne to study, I rebuilt it in stages. First came the upgrade from Java 8 to 17 and the security basics, like secrets sitting in the code and weak password hashing. Then the back end became a REST API and the front end was rewritten in React. Each stage went live separately, so the site stayed up throughout.

Studying here shaped what came next. I'm moving the infrastructure to AWS in Sydney with what I learned in my cloud computing unit: nightly backups already go to S3, and uploaded files are now written to S3 as well. Since most new visitors read English, I translated the interface and later made English the only language. Reminders run on Melbourne time.

Along the way I also added an NBA stats section with 50 seasons of data, open to visitors.

Working out what to build

New features start from what people ask for or run into, and I look into each request before building anything. When people wanted to delete a comment without losing the replies under it, I made deleted comments leave a placeholder so the thread still reads. People also told me the data tables needed too much sideways scrolling on a phone, so I reworked the columns until a phone screen showed seven instead of about four.

The newest piece is the recommended feed on the homepage. It gathers candidates (new posts, popular ones, picks from group owners, people you follow), ranks them with a hot score that halves every seven days, then spreads topics out so one busy group can't take over the page. Each card shows why it's there, private groups never leak into it, and posts you've already opened move down.

A wildlife observation platform that runs on AWS and Google Cloud. People upload photos and videos of animals. The system finds each animal, names the species and stores the count, so you can later search for, say, three kangaroos. Anyone following a species gets an email when it turns up.

Leading the team

I was team lead for four people. I split the work by module, so each person had a clear part, and wrote the architecture and API documents we all built against, including how the two clouds connect.

Where two people's work met, we agreed the details before writing code, such as file names in S3, database fields and the shape of each API. Every pull request needed my approval before it reached main, which kept the parts fitting together.

On the build side I took the highlighted parts of the diagram: the upload with its duplicate check, the species tagging, search by photo, and the React app's structure and login gating. Teammates built the video processing, the database design, tag search and editing, and most of the page designs.

When the cloud lab was wiped

Close to the deadline our cloud lab was reset and everything in it was deleted. I organised the move to a new AWS account and rebuilt the API Gateway and the container functions there. Then I tested the whole system end to end before the demo. We talked a lot and helped each other through the hard parts.

It also showed us which parts we had scripted and which only lived in console settings. Next time the whole setup goes into code from day one.

The code stays private under course rules.

A FastAPI service that spots plastic bags, bottles and other waste in photos, using a pre-trained detection model. I set up a small Kubernetes cluster for it on Google Cloud and load-tested it as I added pods.

Throughput roughly doubled every time the pods doubled. Adding users past that point only made people wait. The limit was the model itself, running on one CPU per pod.