SBHacks 2025: We Won Something!
During SBHacks 2025 we won best entertainment app with Baddy Buddy: Advanced Analytics for Badminton.
SBHacks11 happened during the first week of winter quarter. So far, my 24 units has actually been really fun; it's a lot of work but it's also interesting, and I have a lot of time to spend in the library. My team for SBHacks was a guy I met from Mountain View, my friend from high school, and one of his friends. We decided early on that we'd make something related to badminton and AI, since our friend from Mountain View was actually a top 20 ranked player nationally!
TL;DR
We won 1st place in the entertainment track and you can see the results here.
lce4113Before the Hackathon
Our planning phase was actually pretty tragic. We couldn't find any working models for badminton and we didn't have the compute necessary to train our own. We even considered switching the idea entirely to ping pong (but that would feel like cheating because I did that project last quarter).
Luckily, we kept with the project. I think the one big thing going for it was the name—Baddy Buddy.
After a few library sessions and some last minute scrambling over dinner and at the lounge, we were ready (although Ishan decided to stay up all night until 4AM!).
Hacking Begins
I honestly don't remember much of the first few hours, but nothing crazy happened anyway. Pretty standard arrangement for a hackathon.
We spent some time working with a model called "Widely Accepted Sports Baseline (WASB)" or something like that, but 4 hours in and we decided that it didn't fit our use case. That's when I stumbled upon the holy grail and mighty savior of our project:
sunwuzhou03It basically gives you pose, court, and player detection all in one (it packages all three models nicely in a easy-to-work-with format).
The only problem is that it take FOREVER to run inference on our MacBooks. So at this point, we just stuck with cached results for the sample videos so that we didn't need to recompute.
Flask Backend
We mostly left the SoloShuttlePose repo intact (except for some bug fixes thanks to Omar), but we added our own Flask server to handle web requests for uploading and fetching data. This wasn't to hard, but if I remember correctly we ran into a bunch of problems with importing code from different folder structures (like from a submodule, for example), so we just invoked the python scripts on a subprocess. Actually, this is one of the main reasons SoloShuttlePose saved us; it bundled all three models into one project so we didn't need to manage three venvs and have imports all over the place. This is something I want to learn for the future; how to cleanly mix projects together in python.
Side Quests
There were several mini games that we could compete in to win red tickets, which gave us a shot at the lottery for an electric scooter. I totally wanted that, so I participated in several of them, including a company talk, 1 AM push up contest, 3 AM yoga, and probably some more. We also tried doing a Smash Bros tournament but the wait was too long.
We... Did it?
After about 18 hours in, we finally had a working prototype! Now, it was time to polish the UI and create a full user demo from start to finish. We spent the rest of the time polishing this part up, as the demo is probably more important than the codebase itself. I also came up with a banger video idea and we filmed that. Unfortunately, the judges didn't even look at it because we had to do live demos. But hey, you can look at it... https://www.youtube.com/watch?v=J8pcAyOfyFE
And then, somehow it was morning again. My body physically felt horrible, but I was excited for the judging to start.
Live Demo
Our first judge was the Dean of Engineering. He loved badminton; not so much our app. We should've told him that Ishan was top 20 in the nation, cuz that would've sparked more conversation with him. He didn't even stick for our gen AI feature. Oh well, lesson learned—we gotta be more efficient.
Next we had Prof. Matni. Super chill guy. He liked the presentation, and we executed well. In class the next day, he even told me that "Baddy Buddy" was a hot topic among the judges. Perhaps we could've won one of the grand prizes if we had gotten a little luckier.
The last two were also better presentations, though the 3rd was for sure our best. The dude was so impressed with the vision model integrating with the gen AI model and asked so many questions, and we gave pretty good answers. Even the SBHack student admin who was with him was super impressed.
Overall, we gave good presentations that were an accurate representation of the work we did in the past 24 hours. Unlike some previous hackathons, we didn't sell.
Prize Time!
I fell asleep during the judging period, but woke up just in time for the award ceremony. I had that same fever dream feeling that I had back at GunnHacks where I felt like the work I did was completely garbage and there was no way I could win anything. My team looked at each other knowing that we would probably not get prizes for our project.
BUT WE WERE WRONG: "And the winner for our entertainment track is Baddy Buddy!"
I was so excited. Even though it wasn't a particularly great prize, just the fact that our expectations were exceeded gave us the rush of happiness. Our hard work was validated.
After the Hackathon
Afterwards, I ate lunch at DLG, did laundry, and passed out on my bed. Then I woke up and got ramen with Mason and Surya. So yeah, pretty good weekend.
A few days later, we were featured in the SBHacks Instagram, and it's their most viewed reel of all time so far. Not sure if anyone knows who we are, but it's still cool to be a little "famous" in the coding community. I even had someone go up to me and be like "oh yeah, I think I saw you guys on Instagram." So that's fun.
https://www.instagram.com/reel/DE0S58vyD0W/
Conclusion
Anyway, after a tough loss at GunnHacks last year, it feels refreshing to win something. Looking forward to winning for real next time, and maybe at an even bigger hackathon...