Minus five degrees outside. Misted windows inside. Young people with laptops under their arms crowd the entrance area of the Delta Campus in Berlin-Neukölln. Nobody willingly waits in the cold. Inside it’s about something different: 36 hours of time, hardly any sleep – and the hope that an idea will become more than just code.
No line has been programmed yet – but the conversations in the room reveal: The AI hackathon at the Delta Campus Berlin has begun. Voices overlap, names and LinkedIn profiles are exchanged.
The crowd finally spreads across the area: round tables, long group tables, sofa corners with deep cushions – the location for the next 36 hours.
Laptops open, cables are searched for, jackets are hung over the backs of chairs. A breakfast buffet is set up in the open kitchen: croissants are piled high in baskets. Further conversations arise between two sips of coffee: Who can do what? Eyes keep wandering around the room, looking for possible comrades-in-arms. Small groups form, break up and find each other again.
The organizers received 600 applications for the original 90 places. Among those who have made it are engineers, programmers, students, startup founders and those who want to become one. They are not only coming to Berlin this weekend from all over Germany, but also some from London, Brazil or are participants in Erasmus programs. Therefore, English is spoken as usual.
Read too
Of the 105 participants, there are only about ten women – an imbalance that is noticeable without being spoken about. But typical for the tech scene in Berlin, as organizer Bela Wiertz from Tech Europe says.
Many of the participants already know each other – are friends, fellow students, co-founders or have met at other hackathons.
Clear rules of the game
The agenda starts at 10:30 a.m.: The sponsors introduce themselves and organizer Bela Wiertz explains the rules of the AI hackathon:
- Up to five people can form a team, but lone fighters are also allowed.
- A new project needs to be started. Working on existing projects is not permitted!
- Three sponsor tools/technologies must be used.
The challenge is deliberately kept open: create something new, topic free, no predetermined problem. The catch – or the appeal – lies in the tools: Since three out of four technologies are to be used, there is not much choice or scope. You can choose from Chatarmin, an AI solution for customer service and marketing in e-commerce, the Frontier models from OpenAI, the no-code tool Lovable for creating apps and websites via chat, and Tower, which provides AI agents with reliable data. What emerges from this is in the hands of the teams.
Read too
Not everyone in the room has worked with these programs before. That’s why there are compact introductions, practical instructions and contacts with a deep understanding of technology. Experts go from table to table or can be reached via Discord and answer questions.
“We’re still looking…”
Anyone who is still without a team gathers together for matchmaking. You introduce yourself. Some already bring very concrete ideas, others only a rough direction. “That could fit” – this is how other groups can be found. Because: Nobody wants to stay alone.
The hackathon is not just a competition for the best result, but also a learning space: How does a team form under time pressure? How do you agree on a common idea? How do you think about a solution together – and not just side by side?
Post-its, flipcharts and co.
By lunch, the teams have already been established: people work at the group tables – heads bend together, procedures are discussed, tasks are distributed. Post-its wander across table tops and initial drafts are created on flipcharts.
The burritos at the food counter are picked up quickly: a handle, a nod, back to their place. The food is on the side – the project is not. The focus is clearly on what is to emerge in the next few hours.
The groups are diverse: different levels of experience and generations work together. The participants’ ages range from 18 to over 40, and they are all united by the drive to win the AI hackathon.
The group around Ridvan Sibic has a fixed plan: find an idea that makes sense for the sponsor Arbio. So start a conversation, ask questions, understand what is important – in this case in the context of travel accommodation.
Step two is technical: Which of the assigned tools makes sense? Then the key questions: What’s fun? What would you use yourself? What is annoying in everyday life? Where does the wow effect arise? The aim is not maximum complexity, but rather enjoyment of the product. It can be easy as long as it works.
Then it’s time to build: the idea becomes an agent. Ridvan takes on the role of the most experienced, steers, prioritizes and sometimes brakes. Expectations are deliberately lowered and goals are set realistically, he explains. “The next Microsoft doesn’t have to be created here.”
It is now evening, 9 p.m. The sun has long since set and the Delta Campus is still brightly lit. Empty pizza boxes are stacked in the corners and the smell of tomato sauce still lingers. The first participants made their way home or to the hostel. To sleep so that I can be fit again on day two. There are no overnight accommodations on the Delta Campus.
Around 40 people are still there late Saturday evening and working: codes run across the screens, you can hear quiet typing on keyboards. It has become quieter, the lively conversations from lunchtime have quieted down. Some groups have retreated into glass boxes to work undisturbed. Whoever is still sitting here, working, thinking, tinkering, trying to make the most of the remaining hours.
Coke and mate bottles, empty energy cans and coffee cups are piled on the tables – caffeine gets those who remain through the night.
Done for the deadline
Early on Sunday mornings, the room on the ground floor of the Delta Campus fills up. But not everyone came back.
While most of the participants seemed reasonably rested, around 20 of them worked through the night and only slept on the couch for a few hours – nothing more. One or the other puts their head on the table and falls into a microsleep over the keyboard.
All projects must be completed by 2 p.m. Sunday is final spurt time: Videos are now being recorded in which the teams present their ideas and solutions. Each video can last a maximum of two minutes – that is, precise, to the point and without excesses. After that, everything is in the hands of the jury, which is made up of the hackathon’s sponsors.
Shortly before the deadline there is noticeable tension in the air. Keyboards clatter, mouse pointers fly across screens, groups discuss the last details. Many people work until the very last minute and only submit their projects shortly beforehand.
“Did you submit?”
And then: relief. Pats on the back, high-fives. Pride mixes with exhaustion. But: Not everyone finished on time. Not everyone submitted their project.
There is now an hour to bridge until the finalists are announced. The jury withdraws – a total of 27 projects have to be evaluated. Creativity and technical complexity are assessed. Bonus points are awarded for the effective use of partner technologies.
Some people use the hour of free time to get some fresh air and take a quick trip to the kebab shop on the next street corner. Others look for distraction in the ping pong tournament.
The finale
The finalist round finally starts around 3:30 p.m.: three teams are called and each have five minutes to present their hackathon projects on stage in front of everyone. There is loud clapping and appreciative nods.
The race for the main prize is ultimately made by team “MedAccura” led by Tim Schwarz, a doctor who knows everyday life in clinics very well. It is his second hackathon.
His idea: develop an AI agent for health tech. The problem: Every day, doctors encounter a large number of patients with a wide variety of illnesses and previous illnesses, he says. At the same time, they must constantly be up to date with the latest medical guidelines – PDFs that contain several hundred pages, explains Schwarz. A challenge that is almost impossible to overcome in stressful everyday hospital life.
MedAccura’s solution: a system into which all guideline PDFs are fed so that doctors can access relevant passages via quick search and using OpenAI/ChatGPT.
The jury praised: “This project is characterized by great technical depth and was implemented in a very short time. It has achieved truly great results that can make the world a better place, especially hospitals.” The team wants to continue to pursue the project and meet again after the event to check whether their solution could work in practice.
With these words the hackathon officially ends. The jury concluded: “We are very impressed by all of you.”
