我們如何在一行程式碼都沒寫的情況下贏得黑客松（下）：AI 評審、簡報與冠軍
How We Won a Hackathon Without Writing a Single Line of Code (Part 2): AI Judging, the Pitch, and First Place
Team Null. 憑「Menu of the Future」贏得黑客松冠軍；作品在 Grok Bot 的 AI 評審中獲得 13 分（滿分 15 分），暫列第一。人類評審環節因設備問題未能進行原定的 Grok Bot 即場示範，團隊改以投影片完成三分鐘簡報，最終仍獲冠軍。
Read Part 1 · Original Japanese version of this post
15:30: Getting the submission ready for AI judging
Judging had two rounds: an initial review by Grok Bot, based on what we submitted, and then a pitch in front of human judges. To get through that first round, I had it create a Markdown README that Grok Bot could easily read. I had a rough idea of what the judging bot would be looking for, so I made sure that stuff was in there.
We also tried the novel approach of importing our presentation slides back into the GitHub repo's README. Grok Bot made all of that, too. In the end, no human wrote a single character of code or documentation in the repository.
Here's what we paid attention to:
- Keep unnecessary stuff out of the GitHub repo we're submitting, so it doesn't get mistaken for prompt injection.
- Structure it to use as little of the judging bot's context as possible.
- Make it clear that we have an app that actually works.
- Clearly separate what's implemented and working from what might be possible in the future.
Submissions with prompt injection won't be judged!!
N (
@_I__l__l__l__I_), October 11, 2026. #GrokBotTokyoHackathon
The official deadline was 16:00, but our team's deadline was 15:30, giving us a 30-minute buffer. The submission form was a Google Form, so I had Grok Bot read it ahead of time and set things up so it could fill it out automatically in Chrome inside its VM. Because leaving that to the last second is how you die.
When the 15:30 alarm went off, we started getting the submission ready. Grok Bot's VM filled in every field. It wouldn't click Submit, though, so I took over the browser just for that bit. We high-fived as a team, then hit Submit. Done.
The form closed at 16:00, and Grok Bot's AI judging took about 20 minutes. We'd submitted five minutes before the deadline, so we could just sit back and watch.
The result: Menu of the Future scored 13 out of 15 in the AI round: 5 for the idea, 3 for value, and 5 for scale. We were alone in first place on the provisional leaderboard.
The AI judging results (@GOROman)
“Made it through the bot judging in first place! Next up, the main round!”, Limo (@vibe_walking)
16:30: Trouble during the actual pitch
After a break of about ten minutes, it was time to pitch to the human judges. The nice thing about being first on the provisional leaderboard was that we got to present last. You can see what the other teams are doing, get a feel for the room, and even see which jokes land and which ones bomb. It makes things a whole lot easier.
Eleven teams presented, and then it was finally our turn. That's when things went wrong. We'd had the MacBook Neo and iPad (USB Type-C) perfectly prepped, but when we connected them to the venue's HDMI, nothing showed up. Of course it happens during the actual presentation. On top of that, some mysterious macOS bug meant the screen-mirroring button wouldn't respond at all. We'd rehearsed the demo until it was perfect, so this was a pretty bad spot to be in.
We plugged in N's MacBook Pro instead to show the slides, and mirrored the iPad through QuickTime. But the all-important live Grok Bot demo was no longer possible. So our three-minute pitch was slides only. We talked through how we'd gotten Grok Bots communicating with each other, all the way through to a working mock restaurant on an iPad.
What we'd actually planned to show live was this: place an order through Grok Bot on Limo's iPad, have it arrive at my Grok Bot, and let the organizer bot pull the orders together. Oh well. Let's call it all's well that ends well.
Things I keep in mind when presenting:
- Make the text on the slides BIG. A lot of people use tiny text, and honestly, you can't read it.
- Don't make the audience read the slides.
- Hold the mic and keep it close to your mouth. Or just talk loud enough that you don't need one.
- Look at the audience.
It was hectic, but I think we got across both what we wanted to do and what we'd actually managed to build.
Evening: The results, then the afterparty
As the awards were announced one by one, Team Null.'s name just wasn't coming up. Somewhere along the way, I became convinced: “Okay, we're getting first.”
And we did. First place. I lucked out with a great team where everyone could speak freely.
“🏆 Got first place 🏆”, NULL=RUN (@GOROman), with @vibe_walking and @_I__l__l__l__I_
“We won! Thanks to the best teammates!”, Limo (@vibe_walking), with @GOROman and @_I__l__l__l__I_
What we implemented and how the work progressed. We got a lot built!, Limo (@vibe_walking)
We wrapped up with an afterparty at a nearby izakaya, then headed to Saizeriya for round two. Naturally, we ordered through AI agents. For us, “Menu of the Future” was already “Menu of the Past.”
The end.
I also run a subscription on X. If you'd like to support me, you can subscribe here.
來源：DEV Community · dev.to