This year I swapped my talk for a hackathon

Every year at Kelsus we do a one-week trip called Kelsus Camp. The destination changes each time, and this year we ended up in Foz do Iguaçu, Brazil. It’s a week to spend time together, step away from the remote routine, and create spaces that don’t naturally come up in day-to-day work.

During camp we usually prepare ~40-minute talks on topics we find interesting. As expected, over the last few years artificial intelligence has been taking up more and more space on the agenda.

But something had also been happening to me: the talks were starting to bore me a little. Not because the topics weren’t interesting, but because I noticed the ones I enjoyed most were the ones that included some kind of hands-on activity, even a simple one. When there was something to do, discuss, or build, I stayed much more engaged.

That’s why this year I decided not to give a talk. Instead, I proposed organizing a hackathon and called it the Kelsusthon.

The idea was simple: shake up the camp dynamic for a few hours. Instead of sitting down to listen to another presentation, I wanted us to get into teams and build something concrete, with limited time and a demo at the end of the day.

I also wanted the activity to pull us out of the technical mode we usually operate in. The challenge wasn’t just to code something, but to think up a product idea, make fast decisions, prioritize, put together a pitch, and defend it in front of a jury. And along the way, work with people we don’t always share our day-to-day with.

There was one thing that gave me some doubt: the success of the Kelsusthon didn’t depend only on me. It also depended on the rest of the participants bringing their own enthusiasm and energy. That created some uncertainty for me, but, to my surprise, everyone got really into it and brought out their competitive side — always in a healthy way. Some got so committed to their project they even skipped lunch.

One of the first decisions was to make the hackathon product-oriented. We wanted to step back from our more technical mindset and challenge ourselves to create something creative, useful, and usable — without worrying too much about whether it was perfectly implemented. We also wanted to encourage the ability to “sell” a solution: stand in front of everyone and explain what problem we’re solving, more than how we’re solving it. And we wanted to form teams with people who don’t necessarily work together day-to-day.

With all this in mind, we decided to form teams of three people randomly. A few weeks before the event, we asked each participant for a random fact about themselves. Then, one day before the hackathon, we gave each person the random fact of a teammate they had to find.

The dynamic worked really well. As soon as we revealed the facts, everyone started talking about it and looking for their partners around the hotel. The idea of revealing the teams the day before was so they’d have some time to chat and start thinking about what product they could build the next day.

On the day of the hackathon we started early, around 9 in the morning. Teams had just a few hours to build whatever they wanted. With so little time, everyone had to make the most of any AI tool available: Claude Code, Codex, Gemini — everything counted. In fact, to reach a solid demo in such a short time, it was almost mandatory to lean on good tools, plan well, and use frameworks like SDD and skills.

At 2:00 PM everyone stopped typing. The pitches began.

Team by team, they came up to the front. Each one had three minutes to explain what problem they were solving, how they were solving it, and show a short demo. This part was extremely dynamic: all 13 teams presented in under an hour. As soon as one team finished, the next was already ready to go up.

The projects were incredibly diverse: from a system that reads Slack messages and detects bug reports to automatically fix them, to an app for tracking tennis scores by voice, all the way to marketplaces and GEO analyzers for websites.

After the pitches came the hardest part: choosing the winners. The jury had to score each team in four areas: technical execution, impact and usefulness, innovation and creativity, and pitch and communication. It wasn’t a simple decision. There were very different projects, demos that surprised us, ideas that were sold really well, and teams that managed to build far more than seemed possible in such a short time.

What prizes were on the line? Smartwatches for first place and board games for second and third. But without a doubt, the most important thing was taking home the Kelsusthon trophy.

First place went to TicketFlow, a platform that uses AI to analyze Slack conversations, detect bugs, important tasks and decisions, and automatically convert them into tickets — and even Pull Requests when there’s enough technical context. Second place went to The Arc, a personal planning engine with AI that turns your vision, anti-vision, and goals into weekly milestones and daily tasks adapted to your reality. And third place went to IdeaMatch, a platform that helps transform vague product ideas into structured concepts and find other people with similar ideas to collaborate or validate the problem.

In the end, it’s safe to say the Kelsusthon was a success. It was a different kind of day — full of energy, good ideas, and a level of enthusiasm that exceeded my expectations. Personally, I’m really happy with how it turned out. I enjoyed organizing it a lot, but I enjoyed even more watching everyone get involved and take the challenge seriously. If one thing became clear to me by the end of the day, it’s that this was the first edition of many more to come.