Scan. Score. Explain. (Hack4Humans x AI)

September 16–18, I took part in the Hack4Humans x AI hackathon by Deutsche Telekom, held in Bonn. My team worked on the “Do No Significant Harm” challenge — we had to develop an MVP of a mobile app that can scan any product (food, cosmetics, a mobile phone) and provide information on how safe it is for the natural environment and for people. The challenge is driven by the EU’s introduction, starting in 2027, of new Digital Product Passports that will include information about ingredients, environmental impact, and human safety.
My team consisted of four young students from Germany, an architect, and an account manager. I served as the product owner and prepared the final presentation.

What went well?

• We built a working application with a full React UI and a Python backend, capable of sending requests to Google’s Gemini LLM.
• The app could use a QR code/barcode to retrieve ingredient data and compute an overall safety rating with a detailed ingredient analysis.
• We prepared a polished presentation with a distinctive product brand, a video, product–market fit, audience positioning, unique features, and a go-to-market business model.
• Incredible team cohesion: even though the most experienced team member had no more than one year of development experience, we effectively split roles across frontend, backend, architecture, and LLM integration.
• Fantastic team energy: we communicated a lot, made videos and photos, and joked around.
• During the pitch we had four speakers, two of whom were women (whereas in many other teams, unfortunately, there was often only one speaker).

One of the judges said during the presentation: “It’s incredible that you managed to do so much in such a short time.”

We took 2nd place in the qualifying round, advanced to the final, and finished 4th out of 30 teams.

What could have been better?

• It would have made sense to focus from the outset on the judging criteria (Human Centricity, Innovation, Technical Execution, Impact & Value for the Business), while we concentrated more on getting the MVP to work.
• I spent a lot of time on presentation design (many people noted the strong branding), when I should have devoted more time to the product itself.

What we used

• lovable.ai for building the frontend and designing the presentation. In the past, AI agents struggled with design; this time I was pleasantly surprised by the new capabilities. I especially liked selecting a specific design element and assigning a refinement task to it, and the fact that Lovable automatically fixes code errors as they appear.
• Claude Code for the backend, frontend–backend integration, and frontend refinements. Claude performs well; green-field prototyping works at a high level. But there are still issues: we couldn’t solve the laptop camera access problem on Windows. At the current state of the models, fixing a bug like that can take a lot of time.

Key takeaways

1. AI has changed how we develop software. A year ago, building such a product in so little time would have been nearly impossible. Prototyping and creating a functional MVP is now accessible even without development skills.
2. AI won’t replace developers. AI-generated code needs validation, and architectural decisions are better made by humans. AI can easily produce simple code; complex code often stumps it.
3. The team matters. Even people who didn’t know each other can find a common vibe and team spirit in a few days. Diversity and the ability to listen and trust each other are crucial. Often the most unconventional ideas from today’s students resonated most with the audience.
4. In-person matters. We’ve grown used to remote work, but introductions and onboarding, goal-setting and brainstorming, presentations and coaching work better live.
5. AI can go down a “rabbit hole.” Like a person, it can keep trying the same failing approach. It’s important to reset the context, look at the task from another angle, and avoid wasting time on a dead end.

I’d like to thank T-Digital Greece and George Dedes for the opportunity to participate in this hackathon. And also my team: Sigrid for valuable coaching advice and strong architectural ideas; Lars for leadership and technical expertise; Dea for a customer-centric UI and an excellent presentation; Daniel for high-quality code and the ability to say, “this is impossible to do within the given time frame”; Franziska Titze for the right questions, team spirit, and humor; and Tina for B2B business expertise and empathy. Thank you — these were three incredibly intense days!