Back in 2024, we were known as dingdongs, competing in the Novice category of TIL-AI. We made it to the finals, confident that our models would carry us to the podium. But reality hit hard — one test case failed completely, costing us 100 precious points and dropping us from 2nd to 4th.
Why did it fail?
We literally forgot to import torch.
Fast forward to this year: none of us could commit full-time, and none of us expected to even make it past the qualifiers. We were only here because Tianshi got poached by the organisers — ironically, because of the import torch story — and we joined to support a friend.
The qualifiers were far from smooth. Errors were everywhere, most of our models hit dead ends, and on the final day, we were sitting at 40th place. In a last-ditch effort, a few of us hopped into a Discord call and made a Hail Mary submission for the Reinforcement Learning component.
Somehow, that single move launched us from 40th to 10th — leapfrogging 30 teams and squeezing into the top 16 for the semi-finals, in what we called the Great Leap Forward.
We thought that would be the end of our story. Instead, it became the start of our fairytale run. We upseeded stronger teams in the semis with better-performing algorithms for our guards and scouts, and ultimately secured 2nd place in the finals.
We didn’t win 1st — but looking at where we came from, the story couldn't have been written better. As we like to say:
“We’ve reclaimed what was rightfully ours.”
In this repository, you’ll find what we believe best represents our journey through this year’s TIL-AI — our code, Dockerfiles, experiments, and everything else in between. Our full writeups are available here as well.
To the brilliant and dedicated team:
Brian Hu @BrianHuBuyan
Neo Souw Chuan @neosouwchuan
Tew En Hao @tewenhao
Wong Zenwei @zeotheburrito
To Qi Tianshi @qitianshi, for relentlessly pushing us to join BrainHack again.
To Ryan @ryan-tribex, Ada @HoWingYip, Tianshi, and the amazing teams at AngelHack and DSTA — thank you for building such a fantastic event.
And to the open-source communities behind the models and frameworks that powered this project — we couldn’t have done it without you.