04 | 2026 Kipinä Software

Creative people need to pause every now and then and make room for experimentation. The best way to understand new tools is to use them to freely explore the things that truly interest you.

Kipinä’s Friday hackathon was four hours without specs, a backlog, or a predefined problem. Just an opportunity and a curiosity to see what happens when we combine years of expertise with AI tools.

Read about the ideas, solutions, and insights that emerged from the hackathon!

Insights from the AI Hackathon

Four hours is a short time, but it’s enough for validation—to see if the idea works, if the technology is feasible, and how the AI tool behaves when faced with certain types of problems.

Expertise remains the deciding factor.
The best results come from a deep understanding of the problem. Market analysis is only useful when you know what questions to ask. A security report is valuable only when you know how to interpret it. Artificial intelligence speeds up thinking and action, but it does not replace the ability—born of experience—to narrow down, prioritize, and make meaningful decisions. Without this, AI easily produces only quantity, not quality.
Proper planning pays for itself.
A clear structure before implementation saves more time than any single tool. When a problem is well-defined, AI also performs better: prompts become more precise, the need for iteration decreases, and the end result improves.
The same pattern emerged at the hackathon: those who took a moment to plan ended up making faster progress and getting further.
The Discovery phase has undergone a radical transformation.
Validating a product idea—which involves market analysis, competitor research, prototyping, and iteration—has traditionally taken weeks. Now it can be done in a matter of hours if the team members already have the necessary domain expertise. This fundamentally changes how quickly we can help our clients test and validate their ideas.
The value of a culture of experimentation is emphasized.
When the barrier to trying new things lowers, the decisive factor becomes whether people dare to try, and whether there is room in the community for failure as well. For us, that courage is no accident; it is the result of years of conscious community-building. The four-hour hackathon produced six different projects, lively professional discussion, and at least one operational model that we’re actually going to move forward with. But above all, it reinforced what we’ve believed from the start: the best insights and learning come when people are given the space to experiment together.

These are the ideas and solutions that emerged in four hours:

From Frustration to Product
Arto and Kari are passionate enthusiasts in a field where current digital tools frustrate users and are buggy. Because they knew exactly what wasn’t working, what users needed, and what kind of solution would actually be useful, they used Claude to complete a market analysis and validation in just four hours: there is, in fact, no comparable product in Europe.
Prototypes, competitor analysis, and product idea iteration worked surprisingly well considering how quickly they were developed. This wasn’t about coding on a whim, but rather the discovery phase of product development—a process that traditionally takes weeks or months, but now takes just hours. The reason is simple: a deep understanding of the problem based on user experience.
AI Analysis for Sports Results Services
Olli, the team’s sports enthusiast and CEO of Kipinä, built an AI analytics tool for a sports results service during a hackathon. The analytics are streamed to users in real time, and the admin side manages user permissions and tracks costs and completed analyses on a per-user basis.
From the CEO’s perspective, the pilot addressed a pressing issue for many companies: how to ensure that AI costs remain under control and that the value derived from its use justifies the expense?
The solution utilized Vercel's AI Gateway and Anthropic's Claude. The approach was straightforward: first, design with Claude; then, implementation using agents.
"Overall, I'd say that integrating AI in a controlled manner is surprisingly straightforward these days," Olli commented.
A space to explore, a place to play
Trying something new doesn’t require a fully formed idea—just curiosity and the freedom to experiment. So our expert Tomas started by chatting with his colleagues about open models, self-driving cars, and what they display on their screens. What could we achieve in the field of global modeling?
In just two hours, they created a stunningly functional system called RobotMirror: a real-time 3D avatar whose joints track the user’s movements in the browser, gesture recognition, serverless video chat, Guitar Hero-style multiplayer over the internet, and guided exercise breaks set to a synthesized beat. As a final touch, the lever that adds the user’s face to the 3D world was named: Digital Realities ↔ Added Humanity. Just like the Spark slogan says, of course!
“I was surprised at how quickly we got this up and running. Many research topics that used to be difficult to access are now available as libraries. The software doesn’t have any databases, security measures, or user accounts, so it’s not a very complex system. But developing it was fun, and the test users had a great time trying it out during the hackathon presentation,” Tomas says.
Information Security Audit: The local model identified deliberate design choices based on actual problems
Our expert Ville came up with an interesting experiment: a security audit and critical assessment of code quality for his own project, using a local model (Gemma 3-based, 26 billion parameters) without cloud services—on Kipinä’s Monolith machine.
The result was a pleasant surprise, as the model was able to distinguish between conscious design choices and actual problems. According to Ville, the code quality report was "much better than expected and delivered quickly." The comparison was clear: GitHub Copilot, which is used in the customer’s environment, is often slower, and at best just as fast.
The report was valuable precisely because Ville knew what to ask. He knew how to guide the model to the right areas and identify which findings were significant. Without that expertise, the report would have been nothing more than a long list of technical jargon.
AI-assisted work also requires patience
CTO Jari Huilla built a browser-based clone of *The Incredible Machine*, a 1990s classic puzzle game in which players construct Rube Goldberg-style contraptions.
An ingenious solution emerged during the process: Jari built two custom tools for Claude—one that allowed Claude to independently control the image generator to create sprite graphics, and another that allowed Claude to play the game and take screenshots on his own. The latter was used to fix the levels: of the ten generated levels, only the first one worked as-is.
After about 2.5 hours of work, three out of ten levels were playable. The AI is good at efficiently repeating familiar patterns, but *The Incredible Machine* requires the level design and physics engine to work together in a way that doesn’t happen automatically. The more original the concept, the more human guidance is needed.