We Hacked a Free Tool at the Fastest Hackathon in Berlin
Today I was at the Fastest Hackathon Ever in Berlin. Co-hosted by OpenAI and Langfuse, the format was simple: $50 in API credits, a room full of engineers, a few hours, and a hard demo deadline. No pitches, no planning. Just build something.
I built something that's been sitting in the back of my head for months, pushed there by the same complaint Webentity users keep making.
The problem
Founders spend days rewriting their homepage. They still can't tell if the message lands. And it's not just about whether a human buyer gets it anymore.
When someone asks ChatGPT or Perplexity to recommend a tool like yours, the AI reads your site and decides in seconds. Most startup homepages fail that read. The copy is fine. The problem is structural: the five questions a buyer needs answered are scattered across pages, buried in feature lists, or answered out of order.
What I built: Can-AI-Explain-You?
It's a free MVP at can-ai-explain-you.vercel.app. You paste in a URL and describe your target buyer. The tool crawls your site and audits it against five core buyer questions: what is this, who is it for, what do I get, why should I trust you, and what do I do next. It scores the site on clarity, trust, CTA, and AI-readability, then generates a homepage patch with a before/after preview you can use right away.
The AI-readability score is the part that surprises people. If an AI assistant can't summarize what you do in one sentence, it won't recommend you. Most sites fail this. The patch gives you something concrete to fix, not a vague note to "improve messaging."
Why this came from Webentity users
The same pattern kept coming up. A user would optimize their site for Google, get the design right, write clean copy. Then a prospect would say "I wasn't sure what you actually do."
That's not a writing problem. It's a structure problem. The five buyer questions exist on the site, just not in the right place or order. That's what the tool audits.
What happened at the hackathon
The event opened with a 15-minute session from an OpenAI engineer on using Codex to ship fast, then a short segment from Langfuse on agent tracing and evaluation. After that it was heads-down time.
The room had good energy. Everyone building something real, engineers from both teams walking around to help. At the end, everyone demoed.
Building under a hard deadline is useful. It cuts the gold-plating. You focus on the one thing that makes the tool worth using, and you ship that.
Try it
The tool is free. Paste in your URL, describe your buyer, and see what they actually understand about you.