Zero One Hackathon 🏆
Spots when someone is about to abandon an insurance signup and steps in with a personalized nudge before they leave. Built for UNIQA at a 36-hour AI hackathon.
Role
Built in a team of three over 36 hours at Zero One Hack Vienna.
Challenge
We had to read hesitation from behavioral signals as it happened and answer with the right nudge, like explaining a price or offering an advisor, fast enough for it to still matter. With no real users at a hackathon, we also had to simulate user behavior at scale before we could test any of it.
Outcome
Third place and €500. We simulated users with a mix of Markov chain personas and LLM-driven personas running on Ollama, which kept iteration fast. A Next.js frontend sits on a Python backend that handles signal analysis and LLM reasoning.
Technologies used:
