From hackathon MVP to full-time startup, Germany-based team Trademind (now Kivo Games) took home the Gaming prize at KXSB AI HackXelerator with a fresh take on AI in gaming. While most game AI applications lean on visuals or NPCs, this duo built something entirely different: a recommendation engine that analyzes player behavior and suggests helpful in-game items.

We caught up with Oscar Kerscher, the project’s Tech Lead, to learn more about the journey.

From idea to impact: How Trademind came to life

“We’re hackathon addicts,” Oscar laughed. “Franzi and I have been competing together across Europe for the past six months. When the HackXelerator organizers reached out, we jumped at the chance – 20 days is the perfect pressure cooker for creativity.”

Unlike many AI-in-gaming projects focused on visuals or procedural generation, Trademind flipped the script. “We’re gamers. We know how easy it is to get lost in open-world games with too many options. So we thought – what if we could recommend in-game items based on player behavior?”

The idea started as a behavioral analysis tool for Minecraft. From there, they added a tips-and-tricks generator and a recommendation system for in-game goals and purchases – using AI to help players get unstuck and progress more enjoyably.

AI stack and infrastructure

Kivo Games built their system using a mix of open-source tools and local infrastructure:

• Modeling & pipeline: XGBoost tuned specifically for Minecraft behavioral data, with a locally hosted DeepSeek model.
• Languages: Core code in Python, Minecraft mod in Java.
• Infrastructure: Hosted everything on Oscar’s home server, running Ubuntu and mimicking cloud behavior with remote access for testing and training.
• Dev tools: Cursor helped accelerate early-stage development, though the size of the project later outgrew it.

They initially tried using Restack for their AI agent pipeline but ultimately opted for a manual build.

“We didn’t go with the cloud this time; setting that up quickly in a hackathon environment was a bit out of scope for us. But our home server did the trick!”

The HackXelerator experience

“We got amazing feedback throughout the event, especially from mentors who pushed us to think beyond just building a tool and into building a business,” said Oscar. That push paid off: what started as a recommendation engine is now the basis for a broader, platform-agnostic AI assistant for gamers.

“The end event was the cherry on top. Meeting the organizers and a few other teams in person gave the whole experience a real celebratory vibe.”

What’s next for Kivo Games?

Oscar and Franzi are all in. “We actually quit our jobs to pursue this full-time,” he shared. “We’ve renamed the project to Kivo Games and expanded the scope beyond Minecraft – we want this tech to support any game.”

With acceptance into the HPI Venture Builder, the team is already accelerating toward productization. Stay tuned – they’re just getting started.